Metaphor in Context

Metaphor in Context Josef Stern

A Bradford Book
The MIT Press
Cambridge, Massachusetts
London, England
( 2000 Massachusetts Institute of Technology

All rights reserved. No part of this book may be reproduced in any form by any
electronic or mechanical means (including photocopying, recording, or informa-
tion storage and retrieval) without permission in writing from the publisher.

This book was set in Times New Roman on `3B2' by Asco Typesetters, Hong
Kong, and was printed and bound in the United States of America.

Library of Congress Cataloging-in-Publication Data

Stern, Josef (Josef Judah)

Metaphor in context / Josef Stern.
p. cm.
``A Bradford book.''
Includes bibliographical references and index.
ISBN 0-262-19439-2 (alk. paper)
1. Metaphor. 2. Semantics. I. Title.
P325.5.M47 S74 2000
4010 .43dc21 00-040221
For Cheryl, my Juliet
Contents

Preface xi
Sources xix

Chapter 1
Metaphorical Competence 1 I Knowledge that Metaphor,
Knowledge of Metaphor, and
Knowledge by Metaphor 1

II Metaphor and Context-

dependence: A Quick Tour of the
Argument 9

III Methodological
Preliminaries 21

Chapter 2
From Metaphorical Use to I Meaning vs. Use 36
Metaphorical Meaning 33
II If Literal Meaning, Why Not
Metaphorical Meaning? 39

III Metaphorical/literal Dependence I:

Davidson's Causal Explanation 45

IV Metaphorical/literal Dependence II:

Davidson on Referential Denite
Descriptions, Malapropisms, and
Metaphor 55

V The Autonomy of Meaning and

Metaphor 60
viii Contents

VI Is a Semantic Theory of
Metaphor Possible? 71

Chapter 3
Themes from Demonstratives 77 I Some Prehistory 78

II Character and Content 81

III Direct Reference: Singular and

Predicative 83

IV Indexicals and the Parametric

Determination of Their
Referents 88

V Dthat-descriptions and Complete

Demonstratives 93

VI Demonstrations as
Presentations 95

VII Dthat-descriptions 100

Chapter 4
Knowledge of Metaphor 105 I The Context of a Metaphor:
Presuppositions 107

II Presupposition, Belief, Pretense,

and Presupposition 117

III Utterance-presuppositions and

Metaphor 121

IV Is Knowledge of Metaphorical
Character Really Semantic? 132

Chapter 5
Knowledge by Metaphorical I A History of Similarity in
Content 145 Metaphor 146

II Exemplication 153
Contents ix

III Metaphors of
Exemplication 156

IV Thematic and Inductive

Networks 169

V Lako et al. on Metaphor 176

VI Exemplication, Catachresis, and

De Re Knowledge by Metaphor 187

Chapter 6
Metaphorical Character and I Knowledge of `Mthat' 198
Metaphorical Meaning 197
II Metaphorical Incompetence 201

III The Interpretation vs. Evaluation

of a Metaphor 205

IV Metaphorical Meaning 214

V Metaphor and Indirect Speech

Acts 222

VI Nominative Metaphors 225

VII Metaphor and Simile 229

VIII Is There One Natural Kind of

Trope? 232

IX Three Semantic Theories of

Metaphor: A Comparison 238

X Objections and Replies 248

Chapter 7
Knowledge by Metaphorical I Marie's Problemand Ours 259
Character 259
II The Rise and Fall of Literal
Paraphrasability 262
x Contents

III The Endlessness of Metaphorical

Interpretation 267

IV Metaphorical Mode of
Presentation 272

V Surprise 274

VI Metaphorical Perspective 277

VII Metaphor and Seeing-as 281

VIII Metaphor and Pictures I 285

IX Metaphor and Pictures II 289

X Belief in Metaphor: Marie

Again 295

XI The Moral of the Story 299

Chapter 8
From the Metaphorical to the I Nonlinguistic Metaphors 301
Literal 301
II Historical and Contemporary
Notions of the Literal 303

III Dead Metaphors 309

IV Literal Interpretations as
Context-independent
Interpretations 316

Notes 321
References 363
Index 379
Preface

The last twenty-ve years have witnessed an explosion of books, anthol-

ogies, and journal articles on the subject of metaphor. Exactly what
ignited this intellectual outburst is anyone's guess. It has gone o not only
among philosophers, whose fascination with metaphor was rst sparked
by Aristotle, and among literary theorists, metaphor's native consumers,
but on every intellectual front: among linguists, psychologists, anthro-
pologists, historians of science, art historians, and theologians. What was
once a specialized topic in rhetoric and poetics has now also come to be
fertile ground for interdisciplinary research. Yet when we survey the
plethora of ospring of all this crossbreeding, it is dicult not to wonder
about their species. What about metaphor are all these theories of meta-
phor theories of ? What problem(s) posed by metaphor are they trying to
solve?
We might begin by distinguishing two main kinds of interest that have
drawn thinkers to metaphor over the course of its long history. (At this
point I'll retreat to philosophy, the discipline I know best, but most of
what I say also applies elsewhere.) The rst of these interests is older and,
since the Romantics and Nietzsche, it has also achieved a certain promi-
nence. For thinkers of this persuasion metaphor is an ``intrinsically inter-
esting'' phenomenon, something on the order of human love or a complex
moral problem or the origin of the physical universe, subjects that do, or
should, command our direct attention. Some philosophers nd this in-
trinsic interest in metaphor because they take it to be exemplary of human
creativity, or the fundamental mode of expression in thought and lan-
guage, or a window into the imagination. Others view a metaphor as the
basic unit, or work, of art or of poetry, and still others as a central tool of
scientic explanation or as an essential element of theological discourse.
In short, for everyone with this rst kind of interest, metaphor belongs up
there with The Big Questions. Mark Johnson puts the view well: ``the ex-
xii Preface

amination of metaphor is one of the more fruitful ways of approaching

fundamental logical, epistemological, and ontological issues central to
any philosophical understanding of human experience.''1
Thinkers with the second kind of interest do not nd any such intrinsic
philosophical value in metaphor, or at least no more than they nd in,
say, slips of the tongue or hyperbole. These phenomena may raise interest-
ing empirical or descriptive questions but they sound no deep philosophi-
cal chords. Nonetheless metaphor is interesting for these philosophers
(as well as linguists and cognitive scientists) because it bears on other issues
or questions that are themselves intrinsically interesting. Metaphor excites
these inquirers for the same reason ``exotic'' phenomena draw physicists:
because of their admittedly remote but potentially signicant implications
for general explanatory principles that are of primary interest to the eld.
To draw a comparison closer to home, consider the interest that certain
oddly ungrammatical strings hold for contemporary theoretical linguists,
strings like `We try John to win' or `Himself left'. As phenomena in their
own right, these strings should have no intrinsic interest: They are never
uttered and, therefore, they need no explanation. However, against a
theoretical background they stand out in virtue of the particular ways in
which they are ungrammatical. The bizarre ways in which they are devi-
ant have the power to conrm or falsify hypotheses concerning abstract
principles of grammar, which, in turn, do explain the grammatical prop-
erties of strings that are uttered and, therefore, call out for explanation.
In recent years metaphor has assumed an analogous kind of non-
intrinsic interest for philosophersas well as linguists and cognitive sci-
entistsdeveloping semantic theories of natural language. As increasing
attention has been paid to the nuances, complications, and apparently
irregular aspects of natural language, metaphor has been one case that has
tested and tried our ability to give precise, systematic characterizations of
the ordinary notions we use to describe its ordinary functioning, notions
like meaning (or change of meaning), truth, or signicance. Suppose, for
example, we distinguish between an individual's knowledge of language
including his knowledge of semantic rules that assign interpretations to
strings of wordsand his ability to use that knowledge to make utter-
ances with other meanings and eects. On which side of the divide should
metaphor fall? If metaphorical interpretations of strings are not predicted
by specic proposed semantic rules, what conclusion should we draw?
That the fault lies with the semantic rules? Or that the failure is proof that
metaphor does not fall in semantics and language proper? That it is in-
stead just one among many ways of using (or misusing) our knowledge of
Preface xiii

language? The conclusion we draw will obviously carry serious implica-

tions for our background semantic theory. Within such a theoretical con-
text, metaphor acquires important albeit derivative interest.
I do not intend to fall into the trap of trying to classify all philosophers
who have written on metaphor into one or the other of these two cate-
gories of interest. For one thing, the categories are not mutually exclusive.
For another, there is the danger that one might be tempted to identify
philosophers with the rst kind of intrinsic interest as the ``friends'' of
metaphor and those with a derivative interest as its ``enemies.'' As much
as these labels make any sense at all, friends and enemies of metaphor are
to be found among thinkers with either kind of interest. Hobbes and
Locke, two of the greatest detractors of metaphor in its history, no less
than Nietzsche and Coleridge, two of its greatest inators, share an in-
trinsic interest in metaphor. And who is to say whether someone who
takes metaphor to be a use of language rather than a type of semantic
interpretation is a ``friend'' or ``enemy''?2
So long as we heed these warnings not to abuse the distinction, it can be
helpful to know which kind of interest motivates an author if only be-
cause it will put the problems and topics he addresses in perspective. In
this essay, for example, I am guided in the rst place by a derivative in-
terest of the second kind. Throughout the last two decades' abundant
writing on metaphor, one assumption has been typically taken for granted
(although it is occasionally given a supportive argument and, less occa-
sionally, challenged): that metaphor lies outside, if not in opposition to,
our received conceptions of semantics and grammar, semantics in the
classical sense of the Frege-Tarski tradition and grammar as linguists
conceive it as a speaker's knowledge of language. Some authors (e.g.,
Jerry Sadock, Paul Grice, John Searle, Robert Fogelin, Ted Cohen,
Daniel Sperber, and Dierde Wilson) think that we must supplement se-
mantics with theories of pragmatics, conversation, or speech acts in order
to deal with metaphor. Others (e.g., Donald Davidson, Richard Rorty,
David Cooper) think that the signicance of a metaphor being a use of
language is that it resists any kind of general theoretical explanation, se-
mantic or pragmatic; at best, they hold, we can tell detailed but ad hoc
stories for individual utterances of metaphor. And for yet a third group
(George Lako and his school, and Paul Ricoeur), the fact that metaphor
lies outside the purview of classical semantics is one more example of the
poverty of the tradition, one more symptom that what is really needed is
nothing less than a radical revisionor wholesale rejectionof classical
semantics, which, they charge, was framed on the model of literal lan-
xiv Preface

guage with a built-in bias against ``nonscientic'' or ``nonmathematical''

language.3 Despite the many dierences between these views, they all
assume (generally without argument) that metaphor cannot be explained
by or within semantics. By the same token, it is also taken for granted
that metaphor has little if anything to teach us about semantic theory.
One aim of this book is to challenge these two assumptions: I hope to
show how semantic theory can constructively inform our understanding
of metaphor and how metaphor can illuminate semantic theory in general
and the role of context in theories of meaning in particular.
To be more specic, I am concerned primarily with one question: Given
the (more or less) received conception of the form and goals of semantic
theory, does metaphorical interpretation, in whole or part, fall within its
scope? Or in more material terms: What (if anything) does a speaker-
hearer know as part of his semantic competence when he knows the
interpretation of a metaphor? These questions are not entirely new to
discussions of metaphor, but using the theoretical apparatus of current
semantics we can bring powerful and relatively well-tuned explanatory
tools to bear on them. For example, it is often said that a dening char-
acteristic of a metaphor is that its interpretation ``depends'' on its literal
meaning, but what is the nature of that ``dependence''? Is it a kind of
functionality and, if so, is it semantic or pragmatic? And if the meta-
phorical interpretation of an expression ``depends'' on its literal meaning,
then the expression must still ``have'' its literal meaning. In what sense?
And how is the literal meaning that the expression has, even while it is
interpreted metaphorically, dierent from its metaphorical interpreta-
tion or meaning? As these questions are usually formulated, terms like
``meaning,'' ``interpretation,'' and ``dependency'' are left in a vague,
unexplicated, pretheoretical state, making it almost impossible to give
them denite answers. But the pay-o need not be only for metaphor. If
we can show that features heretofore thought to be peculiar to metaphor
are instances of more general semantic regularities that hold throughout
natural language, we can also enrich the explanatory power of our general
semantic theories.
In addition to these semantic problems raised by metaphor, I shall also
try to throw some light on a spectrum of questions that have heretofore
been addressed mainly by writers concerned with metaphor as an intrin-
sically interesting phenomenon, especially questions concerning the cog-
nitive signicance of a metaphor. However, my approach will dier from
that of most previous metaphor theorists who have taken these problems
to be sui generis to metaphor. Instead I shall try to show how, by em-
Preface xv

ploying our semantic theory, these puzzling aspects of the behavior of

metaphors can be given a diagnosis, or description, that does not deny
their distinctiveness, yet subsumes them under the same rubric as other
semantic facts that hold for nonmetaphorical language.
I cannot, of course, simply assume, as if it needs no defense, that
metaphor lies within the scope of semantics. But the best defense is some-
times a good oense. Although I shall address various objections to the
semantic status of metaphor in chapters 2, 7, and 8, my strongest evidence
will consist in the semantic explanations I propose as working hypotheses.
Yet, I should emphasize that, despite my sympathetic stance toward
classical semantics, I do not mean to suggest that I think that the specic
claims of the current semantic theories I employ are the nal whole truth.
Along with (I would imagine) most contemporary philosophers of lan-
guage, I readily concede that we are still at the very beginning of our un-
derstanding of natural language and that our available semantic theories
are far from nished. This is especially true for our theories of demon-
stratives and the semantic treatment of contextthose parts of semantics
that will matter most for our account. To the degree to which my account
of metaphor rests on these notions, it is also no more than a rst ap-
proximation to a nal answer. It would be better, then, to view this essay
not as an attempt to give a theory of metaphor but, more modestly, as an
attempt to ``map out'' the semantic topography of metaphor. Even if all I
accomplish is to locate metaphor relative to some of the other landmarks
of current semantics or, a bit better, if I persuade you that the attempt to
situate metaphor in relation to current semantics is a project worth pur-
suing, this essay will have succeeded.

The proximate stimulus for this book was the 1974 Linguistics Institute at
University of Massachusetts, Amherst where I took a course on prag-
matics with Robert Stalnaker who introduced me both to his own seminal
essays on presuppositions and to David Kaplan's (then unpublished)
``Dthat.'' My debt to the writings of Stalnaker and Kaplan will be obvi-
ous to the reader. Kaplan's work in particular has been a rich source of
stimulation for my own philosophical imagination, and I hope this appli-
cation of his semantics for demonstratives to metaphor will be a small
contribution to his own program.
It was around the same time that I noticed the failure of substitutivity
of (literally) co-extensive expressions interpreted metaphorically and,
when I returned to Columbia from Amherst in the fall of 1974, the basic
idea of this bookto treat metaphors as demonstrativesoccurred to
xvi Preface

me. My rst sustained attempt to work out the idea was my 1979
Columbia dissertation, written under the direction of Charles Parsons,
Sidney Morgenbesser, and James Higginbotham. I am deeply grateful to
all three of them, not only for their help with the thesis, but for a superb
philosophical education in general. I should also single out Jim Higgin-
botham who rst taught me philosophy of language and linguistics and
then lavished endless hours on the dissertation, generously sharing his
own ideas as well as criticisms. From that same era, I also want to thank
for discussions and comments the late Monroe Beardsley, Merrie Berg-
mann, Arthur Danto, Robert Fiengo, Richard Kuhns, Isaac Levi, Robert
Matthews, Georges Rey, Israel Scheer, Robert Schwartz, Ted Talbot,
and Ellen Winner.
Since arriving at the University of Chicago in 1979, Muhammad Ali
Khalidi, Leonard Linsky, the late Jim McCauley, Ian Mueller, Jerry
Sadock, Joel Snyder, and Bill Tait have oered valuable feedback and
encouragement. A number of conversations with Donald Davidson, who
was still at Chicago when I rst arrived, also helped me appreciate his
position better. I am especially indebted to Ted Cohen for numerous
examples and ideas, some of which are even acknowledged in this book.
A former teacher of mine at Columbia and then my senior colleague at
Chicago, Howard Stein has been a model of intellectual and moral stan-
dards I can only attempt to emulate. According to the rabbis, we learn the
most from our students; in particular I wish to thank Don Breen, Jesse
Prinz, Gabriela Sakamoto, and Lauren Tillinghast for their critical reac-
tions in and beyond the classroom.
Much of the present manuscript was written in Jerusalem, and I have
beneted from the comments of many audiences in Israel and from the
hospitality of the department of philosophy of the Hebrew University. I
am grateful for discussions with Gilead Bar-Elli, Jonathan Berg, the late
Yael Cohen, Asa Kasher, Igal Kvart, Malka Rapaport, Susan Rothstein,
Ellen Spolsky, and Mark Steiner. Sidney Morgenbesser rst told me to
seek out Avishai Margalit while I was writing my dissertation, and in
addition to everything I have learned from his own papers on metaphor,
he has been one of my best critics since then. Another debt I owe Avishai
is that he rst introduced me to the Library of the Van Leer Institute, a
remarkable oasis of philosophical composure in Jerusalem where the
penultimate drafts of this book were composed during 19951997.
During its last stages of preparation, the manuscript beneted from the
criticism of Sam Glucksberg and Boaz Keysar (on ch. 5) and several ref-
erees for MIT Press. In particular I want to thank Mark McCullagh for
Preface xvii

very helpful comments especially on chapters 1 and 3 and Roger White

for criticisms of chapters 2, 5, and 6; both signicantly improved the
manuscript. It wasn't, unfortunately, until very late in the writing that I
learned of White's own recent book on metaphor. Nonetheless I have
tried to incorporate responses to a number of White's concernsas well
as shamelessly drawing on his impressive knowledge of Shakespeare and
literature for examples. My thanks, too, to Amy Brand, Carolyn Gray
Anderson, and especially Judy Feldmann of MIT Press for all their help
and advice in the nal production of this book. Last of all, I have bene-
ted from written and oral exchanges in recent years with Murat Aydede,
David Hills, Michael Leezenberg, Patty Nogales, and Francois Recanati.
I am most grateful to various foundations who translated their faith in
this project into essential material support: the Giles Whiting Foundation
and the Lawrence Chamberlain Fellowship while writing my dissertation
in 19771979; the Lady Davis Foundation, for a postdoctoral fellowship
in 19841985; the American Council of Learned Societies, for a fellow-
ship in 19881989; the Chicago Humanities Institute of The University of
Chicago, for a quarter's fellowship in 19911992; and the National En-
dowment for the Humanities, for a fellowship in 19961997. I would also
like to thank the Division of the Humanities, The University of Chicago,
for its support during my research leaves and especially Stuart Tave,
Dean in 19881989, for his encouragement.
In Hebrew, aharon aharon haviv: Last is dearest. I want to thank my
parents, Kurt Stern and the late Florence Sherman Stern, for their con-
tinual love and support. David Stern, my unliteral identical other half,
constantly mistaken for me just as I am for him, has graciously agreed to
accept full responsibility for all blunders in what follows. From Amitai,
Ra, and Yoni, my own Stern Gang, I have learned how many metaphors
resist literal expression; my thanks to them also for use of the computer in
their spare time. Finally, not only can't literal words express what I owe
Cheryl Newman for the love and devotion that made this book possible
and continue to make most everything else in my life worthwhile; even
metaphor can't say it all.
A nal disclaimer: Any resemblance between characters mentioned in
examples and actual persons, living or dead, is entirely coincidental.
Sources

I wish to thank the following for permission to reprint revised selections

of previously published papers:

``Metaphor as Demonstrative.'' From The Journal of Philosophy 82

(1985): 677710. Used by permission of the editor of journal.
``Metaphor without Mainsprings: A Rejoinder to Elgin and Scheer.''
From The Journal of Philosophy 85 (1988): 427438. Used by permission
of the editor of journal.
``What Metaphors Do Not Mean.'' From Midwest Studies in Philosophy
(volume XVI): Philosophy and the Arts, edited by Peter A. French,
Theodore E. Uehling, and Howard K. Wettstein. ( 1991 by University
of Notre Dame Press. Used by permission of the publisher.
``Metaphors in Pictures.'' From Philosophical Topics 25/1 (1997): 255
293. Used by permission of the editor of journal.
Chapter 1
Metaphorical Competence

I Knowledge that Metaphor, Knowledge of Metaphor, and Knowledge

by Metaphor

Suppose that Romeo and Juliet actually existed and did everything
Shakespeare says they did.1 And suppose that Romeo uttered (1)
(1) Juliet is the sun
in a context exactly like that depicted in the respective act and scene of
Shakespeare's play. What did Romeo say in uttering (1)? How should we
interpret his utterance? And what does oneboth Romeo and his audi-
enceknow when one knows the interpretation of (1)?

A rst stab at describing Romeo's utterance might go like this. Romeo,

we safely assume, knows the dierence between a human and the sun;
hence, he cannot intend to say what his words literally mean: that the
human Juliet is the celestial sun. Instead his utterance should be inter-
preted in some other nonliteral way. What he has said under this alter-
native mode of interpretationgiven what we know about Romeo, Juliet,
and the sun in that contextis, say, that Juliet is exemplary and peerless,
and/or that she is worthy of worship and adoration, and/or that he can-
not live without her nourishing attention. Let's call these ``metaphorical
interpretations'' of the utterance of (1).
Tasks of metaphorical interpretation like this are legion in everyday life
as well as in the more sophisticated contexts of literature and poetry. As
speakers and interpreters we perform them with the same naturalness,
ease, and sense of competent comfortableness with which we interpret
other kinds of utterances. Most (though not all) of the time we succeed in
interpreting even the most novel and imaginative metaphors without spe-
cial diculty (even when they require more eort). Explaining this ordi-
2 Chapter 1

nary achievement, however, is a task of a dierent order for the theorist of

language.
Two comments will help clarify the character of our task. First, let's
abstract away from how speakers and hearers respectively understand an
utterance, the physiological and psychological abilities they employ in
their processing, production, and perception of (metaphorical) speech.
Instead let's focus on what they each know when they understand an
utterance and its interpretation. If and when they succeed in communi-
cating, there must be common understanding, an interpretation they both
grasp. Our task is to identify that common object of knowledge and its
structure.
Second, the word ``interpretation'' in popular usage is used to refer to a
mixed bag of contents; in the case of metaphor, there is additional con-
troversy over the particular contents that belong in the bag. (Indeed this
book might be read as an attempt to argue for one particular kind of
content to be included among those for metaphor.) For now, I shall sim-
ply stipulate what I mean by the term; the proof will follow in the eating. I
demarcate my notion of interpretation along three dimensions.
(i) A metaphorical interpretation is propositional: it is what is said
on an occasion of utterance, its informational content (leaving it open
for now whether this consists in truth-conditions or something else).2
Although many nonpropositional elementsfeelings, attitudes, images,
and other associationsmay also be ``conveyed'' (to introduce a neutral
term) by the utterance of a metaphor, I do not include them within its
interpretation.
(ii) Like any utterance, a metaphor typically conveys more information
than its interpretation; one knows that the speaker is speaking metaphor-
ically, in English, addressing someone, in a tone expressing a particular
emotional attitude, etc. Of these various conveyed pieces of information,
the interpretation is the information that is either semantically encoded in
or determined by the utterance relative to specied contextual parameters.
(I spell out these latter notions in chs. 34.) In more familiar philosophi-
cal terminology, I am concerned with the proposition expressed by the
utterance of the metaphor in its context of utterance. This semantic in-
terpretation is not ``everything'' a metaphor ``means''; sometimes it is
not even what is most interesting about the metaphor. But whenever we
understand an utterance, I assume that a central part of our understand-
ing consists in knowing its propositional contentand the same holds for
utterances of metaphors.
Metaphorical Competence 3

(iii) Among the propositional interpretations of the utterance of a meta-

phor, some but not others may be intended by the speaker. Let's say that
an interpretation that meets the rst two conditions is a possible interpre-
tation of an utterance even when it is not what its speaker intended, had
in mind, or occurrently entertained on the occasion. Of course, when we
ask for ``the'' meaning of a metaphor, we are usually asking for the
speaker's intended interpretation, the interpretation he meant, not simply
a possible interpretation. However, what it is possible for a speaker to
intend by a given metaphor depends on what it is possible to express with
it. It is with all those possible interpretations that I am concerned, dis-
regarding the additional teleological, intentional condition built into our
ordinary usage.
For now, then, let's assume that at least some utterances containing
expressions used metaphorically express possible propositional inter-
pretations. The task of this book is to answer the question: What does a
speaker-hearer3 know when he or she knows such an interpretation of a
metaphor?
This question should be broken down into two subquestions. But be-
fore turning to them, we should distinguish a third question that has been
widely discussed in the metaphor literature although I shall pursue it here
only briey. This is the question of how one knows that an utterance is a
metaphor. What are the conditions, heuristics, clues, cues, trains of rea-
soning, or steps followed by speaker-hearers by which they identify or
recognize particular utterances as metaphors, rather than as literal utter-
ances or as nonliteral utterances of other kinds or as strings of nonsense
sounds? Let's call this the ``recognition'' question.
Once upon a time (and in some quarters still nowadays), many philos-
ophers would have responded to the recognition question with necessary
and/or sucient conditions that signal and even ``warrant'' the judgment
that an utterance is to be interpreted metaphorically.4 The received story
went like this: (i) Every utterance is presumed to be literal until proven
otherwise, that is, unless it is ``impossible'' (in a sense that usually goes
unexplained) to interpret it literally. (ii) A class of utterances M (for in-
stance, our own example (1)), taken at their literal face value, are either
grammatically deviant, semantically anomalous, explicitly or implicitly
self-contradictory, conceptually absurd, nonsensical, category mistakes,
sortal violations, pragmatically inappropriate, obviously false, or so ob-
viously true that no one would have reason to utter them. Hence, each
utterance in M, were it interpreted literally, would be ``deviant'' in one or
4 Chapter 1

another of these ways (choose your own favorite). (iii) Presented with one
of M, the hearer recognizes that it is ``impossible'' to interpret it literally
and (iv), therefore, identies it as a metaphor. (v) M, it is also assumed,
contains all (and only those) utterances actually identied as metaphors.
Despite the venerable tradition of writers on metaphor who explicitly
or tacitly endorse this account, it is subject to fatal descriptive and ex-
planatory problems. Of course, many metaphors happen to be, as a matter
of fact, literally ``deviant'' in one or the other of these ways. But there are
also countless counterexamples to the conditions of (ii), counterexamples
that are exceptional only insofar as philosophers have ignored (or re-
pressed) them for so long. These counterexamples have perfectly good
literal as well as metaphorical interpretations, and some are even ``twice-
true,'' that is, true in the very same contexts both when they are inter-
preted metaphorically and when they are interpreted literally.5 A few
examples:
(a) Mao Tse-tung's comment, ``A revolution is not a matter of inviting
people to dinner.''
(b) The caption on a photo of Japanese nuclear reaction plants (in a
Time article on the pros and cons of nuclear power following
Chernobyl): ``Japan: the land of Hiroshima and Nagasaki feels it has no
alternative.''
(c) A description of a radical reinterpretation of ``Endgame'' for which
Beckett sued the director: ``Audiences today are accustomed to gospel
versions of Sophocles or Paleozoic resettings of Shakespeare. But
Sophocles and Shakespeare live on Parnassus. Beckett lives in Paris''
(Time, Jan. 21, 1985).
(d) ``[H]e was esteemed by the whole college of physicians at that time,
as more knowing in matters of noses, than anyone who had ever taken
them in hand'' (Lawrence Sterne).6
(e) ``Two roads diverged in a wood, and I
I took the one less traveled by,
And that has made all the dierence'' (Robert Frost).
(f ) An article on the merger of the two Wall Street investment houses
Morgan Stanley and Dean Witter begins with the simile/metaphor: ``If
Morgan Stanley is like a bued pair of calf-skin oxfords, then Dean
Witter is a comfortable pair of broken-in loafers.'' This is clearly
gurative (and, as we shall see in ch. 5, a good example of a metaphor
Metaphorical Competence 5

schema or family). However, the following (italicized) clauses in the

same article are metaphorical (or a mix of metaphor and synecdoche)
and literal in the same context:
Those bridges [connecting the two rms] will connect vastly dierent cultures.
Morgan Stanley's senior executives, in their tasseled Gucci loafers and Armani
suits, make many millions of dollars a year as they jet around the world to do
deals with government ministers and business tycoons. Their counterparts at
Dean Witter, meanwhile, favor Brooks Brothers suits and Dexter shoes and
shuttle around on suburban highways, where even the more successful earn at
most several hundred thousand dollars a year hawking mostly plain vanilla
mutual funds, stocks and bonds to Main Street America. (International Herald
Tribune, Feb. 10, 1997)
(g) An example where an expression that could be taken either (both)
literally or (and) metaphorically is to be taken literally rather than
metaphorically: The publisher of the Village Voice, David
Schnidermann, was reported to have said: ``I want the Voice to be a
journalistic player in this town and not a cute little thing from the '60s
that amuses everyone from time to time with its own internal food ghts''
(my emphasis). Since we would naturally take the italicized phrase to
be a metaphor, the reporter immediately added in parentheses: ``He was
not speaking metaphorically. An angry Voice writer once threw potato
salad at the Letters Editor'' (International Herald Tribune, Feb. 21, 1997,
p. 22).
(h) ``Man, after all, is not a tree, and humanity is not a forest''
(Levinas 1990, 23).7
From a simply descriptive perspective, these counterexamples demon-
strate that the received account that relies on some kind of deviance con-
dition cannot be the whole story. Furthermore, even if some version of
the deviance condition were satised, its explanatory signicance is not
obvious. At most the deviance would explain why the sentence is not
taken literally, not why it is interpreted metaphorically, a point that
requires further explanation because a single string frequently admits
alternative nonmetaphorical (but nonliteral) as well as metaphorical
interpretations.8
There is also a deeper diculty with the received story. The deviance
account proceeds on the presumption that every utterance is rst inter-
preted literally. Following a serial or linear model of processing, the
speaker-hearer turns to a nonliteral interpretation only after the literal
interpretation has been eliminated. However, it is equally possibleand
6 Chapter 1

we now possess conrming evidence for this alternativethat both literal

and nonliteral interpretations are processed and evaluated in parallel or
simultaneously. Among such parallel alternatives, the preferred inter-
pretation is the one that is most accessible (where accessibility is itself
context relative) rather than the one that is most literal.9 Such a parallel-
processing model is also indirectly supported by the observation that the
criteria we use to select a metaphorical (or any other kind of nonliteral)
interpretation over a literal one cannot be sharply distinguished from the
varied criteria we use to select among alternative nonliteral interpretations
of one utterance. Just as there is no one condition in the latter case, there
is no reason to think that there is a single necessary and/or sucient
condition that overrules the literal for the metaphorical. Adherents of the
grammatical deviance condition narrowly focus on the limiting case
where the metaphorical interpretation is the only possible interpretation of
the utterance. But their tunnel vision obscures the fact that we may select
one over another interpretation because it is the best (or better) rather
than the only candidate. And judgments about the best X typically depend
on multiple balancing factors rather than necessary and/or sucient con-
ditions. Moreover, among these factors it is not the actual syntactic and
semantic properties of the sentence uttered that matter the most, but the
speaker-hearer's beliefs and presuppositions about those properties as well
as about the purpose and setting of the utterance. These factors make
metaphor identications more like judgment calls with no single determi-
nate answer than like warranted judgments.
Jonathan Culler gives us a good example of the subtle interplay and
unresolved indeterminateness of metaphor recognition on a parallel-
processing model. When we read the sentence in Hamlet, ``Look, the
morn in russet mantle clad Walks o'er the dew of yon eastern hill,'' we
face multiple interpretive possibilities: Either we can assume ``that in the
world of this play (which does, after all, contain ghosts) morn is a gure
that walks over hills; we could posit that Horatio is hallucinating (the
ghost has been too much for him); or we could assume that the morning
here behaves in accordance with our usual models of verisimilitude and
that the false assertion that morning `walks' should lead us to reect on
the qualities of dawn.''10 On the rst two alternatives we take the sentence
literally, as the description of something either fantastic or illusory; only
on the last alternative is any expression in the sentence interpreted meta-
phorically (and there are obviously further metaphorical alternatives). In
``deciding'' on one of these interpretations, or in evaluating them, it
should be noted that Culler does not appeal to any sentence-internal fea-
Metaphorical Competence 7

ture of the utterance (which, as a matter of fact, is deviant); rather, to

varying degrees, he presupposes some (partial) understanding of what the
interpretations of the sentence on its various literal and nonliteral alter-
natives would be. In any case, he concludes, there is no single unique kind
of interpretation that we can condently assign to the utterance once and
for all. Depending on subtle preferences that will vary with the context,
one or the other of the interpretations will be more or less appropriate.
Although we undoubtedly do exploit a wide range of contextual cues in
this task of identication, it is essential to distinguish this role of the
context in selecting among types of interpretations from its other role(s)
in determining the content of an interpretation.11 We can know that an
utterance is to be interpreted metaphoricallyrather than literally or in
some other nonliteral waywithout knowing what its metaphorical in-
terpretation is (or would be), and, inversely, we can know what the
metaphorical interpretation of an utterance would be even when we do
not know whether it should be recognized as a metaphor on that occa-
sion. The competence or abilities involved in the one task should be dis-
tinguished from those of the other. And for the remainder of this book,
I'll be concerned with metaphorical interpretation rather than recognition.
Bracketing the recognition question, it will be helpful at this juncture to
distinguish two subquestions about the interpretation of a metaphor:
1. What kind of knowledge, or ability, enables a speaker-hearer to inter-
pret a metaphor? Is it part of one's general knowledge of language, a
species of one's semantic knowledge, the same competence that underlies
one's ability to interpret nonmetaphorical, ``literal'' language? Or is it, in
whole or part, extralinguistic? And in the latter case, is it a yet-to-be-
identied power that lies beyond the ordinary speaker's repertoirea
kind of genius, as Aristotle and perhaps Kant hinted? Or is it simply one
among the many ordinary (though no less remarkable) abilities we all
possess to use ordinary language in an indenite number of ways?
2. What type of knowledge, cognitive content, or informationif any
does the utterance of a metaphor express or convey? Is it a kind of infor-
mation or cognitive content that can only be communicated, or expressed,
by a metaphor (the same or another)? Or could it be expressed equally
well in literal language?
The rst of these two subquestions addresses, I shall say, our knowl-
edge of metaphor; the second, our knowledge by metaphor. Although
the two questions are distinct, they are frequently conated. Suppose a
speaker's knowledge of metaphorical interpretation is part of his gen-
8 Chapter 1

eral linguistic competence, knowledge that includes a stock of concepts

that constitute the linguistic or lexical meaningsliteral meanings, if
you willboth of the simple words in the language andtogether with
knowledge of their modes of compositionof the complex expressions
that can be composed from those simple words. Then at least ``in princi-
ple''that is, assuming he possesses the literal vocabularyany concept
or meaning he is able to express with a metaphor should belong to the
stock of linguistic, or literal, meanings he knows in virtue of his knowl-
edge of language. Hence he should be able to express any of those con-
cepts using some word(s) literally (even if more awkwardly). Thus the one
position on our knowledge of metaphor would seem to entail a correlative
position on our knowledge by metaphor.
On the other hand, if one's ability to interpret a metaphor is something
other than, or in addition to, his general linguistic competence, it should
at least be possible for the contents of at least some metaphors to be in-
expressible by any literal expression. The extralinguistic ability underlying
metaphorical interpretation might be common or singular but, in either
case, there would be no reason to thinkand, in some cases, perhaps
good reason not to thinkthat it must be possible to put the meaning or
concept expressed by the metaphor into literal words, words that express
concepts or meanings the speaker-hearer knows simply in virtue of his
knowledge of language. Again, the second position on our knowledge of
metaphor would seem to entail a corresponding position on our knowl-
edge by metaphor.
Following these lines of reasoning, our knowledge of metaphor and
knowledge by metaphor go hand in hand. This, in turn, encourages the
view that if our knowledge of metaphor is part of our general semantic
competence, and everything the metaphor expresses is (in principle) ex-
pressible literally, the metaphorical mode of expression is cognitively dis-
pensable. From which it is next argued that, because no literal paraphrase
of a metaphor ever is adequate and the metaphorical mode of expression
is manifestly not eliminable, our knowledge of metaphor is not part of our
semantic competence. In this book I shall argue that this conclusion does
not follow.12 There is an essential component of a speaker's knowledge
of metaphor that lies within his general semantic competencealthough
this knowledge is just a component and is never by itself sucient to
yield knowledge of a metaphorical interpretation (for which knowledge
of context is also necessary). Furthermore, this semantic knowledge of
metaphor is the same kind of competence that underlies a speaker's
knowledge of a signicant subclass of so-called literal language, namely,
Metaphorical Competence 9

demonstratives and indexicals. Finally, it will turn out that this very same
competence enables the speaker to express knowledge, or information, by
a metaphor that is not expressed in literal paraphrases of the interpreta-
tion of the metaphor; indeed it is not expressed except through the meta-
phorical mode of expression of the metaphor.
The arguments for these complementary claims constitute the bulk of
this book. But because it will require some preparatory work to lay out
the account, in the next section I'll present an overview of the theory. This
sketch will hopefully raise the reader above the trees so she can glimpse
our destination despite the winding path through the dense forest that our
argument will follow.

II Metaphor and Context-dependence: A Quick Tour of the Argument

Consider again (1),

(1) Juliet is the sun,
its literal meaning L, and the various things it conveys (again, to use this
as a neutral term) metaphorically about Julietthat she is exemplary and
peerless, worthy of worship and adoration, one without whose nourishing
attention another cannot live, one who awakens those in her presence
from their slumbering, who brings light to darkness. Call these features
M.13 The question at the center of the major disagreements over meta-
phor among philosophers, linguists, and others in recent years concerns
the precise way to characterize the relation between the utterance of a
sentence like (1), its literal meaning L, and the various possible meta-
phorical conveyances in M. Everyone agrees that L isat least if any-
thing isa semantic interpretation of (1), something we know in virtue of
our semantic competence. Everyone also agrees that somehow, by utter-
ing (1), the speaker conveys something metaphorically about Juliet, one
or another of the features in M. Where writers disagree is over the nature
of the relation R between these three entities: the utterance, L, and M. Is
M (or some of its elements) something that the hearer is caused to notice
or to infer as an eect of the utterance of (1) with its literal meaning L? Or
is M something the speaker means as opposed to the (literal) meaning L
of the sentence? Or is M, though not literal, nonetheless an interpretation
determined by the semantics of the language much as is L?
There is no lack of answers to these questions but the vast majority of
authors in the metaphor literature have answered the last question
whether R is a semantic relationwith a resounding ``No.''14 Since we
10 Chapter 1

shall touch on the many reasons for this denial in coming chapters, I shall
not review them all here. However, to motivate my own account whose
starting place is the ``context-dependence'' of metaphorical interpreta-
tion, let me begin with one deeply held source of resistance to semantic
approaches to metaphor.
Suppose our ability to interpret a metaphor were solely a matter of our
semantic competence; then all semantically competent speakers ought to
be able to interpret all metaphors. But not all speakers can. Therefore,
there must be something else to metaphorical competence. Either it is a
special power, like the kind of singular genius Aristotle and Kant may
have envisioned, or it is an ordinary skill of speech that, for lack of a
better word, we can subsume under the umbrella word ``use.'' On either
alternative, the idea is that, unlike standard cases of semantic interpreta-
tion, the interpretation of a metaphor varies so irregularly, idiosyncrati-
cally, and unpredictably that no theory, let alone a semantic theory, could
aspire to explain it. In a pejorative sense, this is also what authors some-
times mean by the slogan that metaphor is context-dependent, in contrast
to literal interpretation, which is claimed to be context-independent:
invariant, predictable, and regularhence within the domain of a theory
and in particular a theory of meaning, as opposed to atheoretical (or
antitheoretical) use. Thus Richard Rorty:
[S]emantical notions like ``meaning'' have a role only within the quite narrow . . .
limits of regular, predictable, linguistic behaviorthe limits which mark o
(temporarily) the literal use of language. In Quine's image, the realm of meaning is
a relatively small ``cleared'' area within the jungle of use. . . . To say . . . that
``metaphor belongs exclusively to the domain of use'' is simply to say that . . . [it]
falls outside the cleared area.15
In this sense, its ``context-dependence'' or status as ``use'' would seem
to render metaphor impregnable to any kind of theoretical explanation
and to semantic theory in particular. Donald Davidson gestures toward
the same view when he groups metaphors together with works of art and
dreamwork, all of which he locates in the realm of the imagination where,
we are given to believe, anything goes: ``understanding a metaphor is as
much a creative endeavor as making a metaphor, and as little guided by
rules.''16 This description, which recalls Aristotle's and Kant's discussions
of genius, might t the creations of some masters of metaphor, but it is a
far cry from the metaphorical competence required for its mastery by the
ordinary interpreter. Indeed the opposite is the case: much more regu-
larity and predictability characterize metaphorical interpretation than the
impression fostered by Rorty and Davidson suggests. First, there is evi-
Metaphorical Competence 11

dence of substantial interpersonal agreement among speakers over the

classication of utterances as metaphorical rather than literal and over
paraphrases of particular metaphorical interpretations (in contrast to
paraphrases of sentences classied as nonsense).17 This argues against the
perception that metaphor is entirely unpredictable and idiosyncratic,
although it does not yet support the view that metaphor is semantically
regular or law-governed. Second, and more relevant for our semantic
claim, when we look beyond the individual interpretations of particular
utterances of metaphors to reoccurrences of the same expression, each
time used metaphorically, in dierent sentences uttered on dierent occa-
sions, we nd a second kind of interpersonal regularity. For example,
contrast the interpretation of `the sun' in (1) with its interpretation in
(2) Achilles is the sun
where it expresses Achilles' devastating anger or brute force; or in
(3) Before Moses' sun had set, the sun of Joshua had risen (BT
Qedushin 72b)18
where it expresses the uninterrupted continuity of righteousness, which,
according to the Talmud (based on Eccl. 1: 5), preserves the world; or in
(4) ``The works of great masters are suns which rise and set around us.
The time will come for every great work that is now in the descendent to
rise again.'' (Wittgenstein, Culture and Value)
in which `sun' expresses the cyclicity and eternal recurrence of greatness,
that things once great will be great again; that descent will be followed by
ascent, by descent, and so on. Or consider the dierent metaphorical
interpretations of `is a bubble' in
(5) Life is a bubble
and
(6) The earth is a bubble19
or the two metaphorical uses of `hill' by John Donne in (7) and (8):
(7) . . . On a huge hill,/Cragged, and steep, Truth stands, and he that
will/Reach her, about must, and must go . . . (Satire 3)
(8) The Church is such a Hill, as may be seen every where. (Sermon 13)
According to the eminent Donne scholar John Carey, the poet's choice of
`hill' in (7) expresses the idea that ``it is necessary to take a circuitous
route, investigating the claims of dierent churches, before reaching [the
12 Chapter 1

truth],'' while in (8) it expresses the contrary thought that the position of
the church ``allows it to be seen unmistakably from all sides, so there is no
need to investigate the claims of dierent churches.''20 Or, nally, com-
pare these two metaphorical interpretations of `is (like) a martini':
(9) A great diamond is like a perfect martinicool and sexy. (Timothy
Green, The World of Diamonds)
(10) The University of Chicago is like a martini. There are some people
who nd it an acquired taste. (Charles O'Connell, former Dean of
Students, The University of Chicago)
Ignoring details, we should agree rst that there is some dierence be-
tween the metaphorical interpretations of the members of each of these
sets. Second, we should observe that these dierences seem to correspond
to some dierence related to a feature of their respective contexts. As a
rst conjecture, we might think the relevant dierence in context is lin-
guistic: the dierent subject noun phrases with which the metaphor, in
predicative position, co-occurs (or, in Max Black's well-known terminol-
ogy, the dierent frames in which the same metaphorical focus occurs).
However, it is easy to see that the same kind of dierence of metaphorical
interpretation can also arise where dierent tokens of one sentence occur
on dierent occasions with dierent beliefs or attitudes associated by the
speaker-hearer with the noun phrase (or frame). In contrast to Romeo's
utterance of (1) in the context depicted in Shakespeare's play, imagine an
utterance of (1) in a context in which Paris's opinion of Juliet is that she is
the kind of woman who destroys admirers who try to become too close or
intimate with her. In that context, (1) might be used to warn Romeo not
to get involved with Juliet. As this example shows, the relevant dierence
in context must include extralinguistic, and nonverbalized, attitudes
(however we work out the details). The moral I draw is this: There may
be little systematic or predictable so long as we look just at the particular
contents of the dierent metaphorical interpretations one by one, but at
one level of interpretation more abstractat a level that relates each
content of the same expression used metaphorically to a relevant feature
of its respective context of use (whatever the relevant feature of context
turns out to be)metaphorical interpretation does seem to follow pat-
terns and to support predictions. Same expression, same context, same
interpretation; same expression, dierent contexts, dierent interpreta-
tions. Thus the degree to which we nd metaphorical interpretation to be
regular and predictable depends on the level we focus on. As we shift our
attention one notch upward, we discover regularities and systematicities
Metaphorical Competence 13

that otherwise go unnoticed. The structure of these variations, as we

shall see in later chapters, is essential to understand both the productivity
of metaphor and speakers' mastery of the mechanism of metaphorical
interpretation.
A second group of philosophers who are no less antithetical to a se-
mantics of metaphor rejects this ``nihilistic'' reading of the slogan that
metaphor is context-dependent. Their resistance stems from a second
source that holds that our knowledge of metaphorical interpretation is
context-dependent or a matter of use in that it is a function not, or not
only, of our knowledge of language but (also) of all sorts of extra-
linguistichence, contextual as opposed to linguisticknowledge, beliefs,
and skills. Metaphorical interpretations are either built up out of our
extralinguistic beliefs, common knowledge, and presuppositionsabout
extralinguistic entities such as Juliet, martinis, the Church, and the Uni-
versity of Chicagoor they are a function of psychological abilities such
as the perception of similarity or analogy, abilities or skills that are not
language-specic and that are employed in all sorts of nonlinguistic
modes of communication or symbolization. These abilities and the facul-
ties responsible for the relevant presuppositions fall beyond a speaker's
semantic competence proper; hence, metaphorical interpretation also falls
outside the scope of semanticsand instead in either pragmatics, theories
of language-use or speaker's meaning, or general accounts of cognitive or
symbolic activity.21
This inference from use or context-dependence to the nonsemantic
status of metaphor cuts to the heart of my own position. I also begin from
the observation that metaphorical interpretations are composed out of
presuppositions and beliefs that are not part of our knowledge of lan-
guage proper, that are instead acquired through general symbolic skills.
What is right about the second view is that metaphorical interpretation is
not exclusively a function of linguistic competence. However, in contrast to
the second view, I shall argue that the extralinguistic context-dependence
of a metaphor is nonetheless compatible with and, indeed, requires se-
mantic knowledge in order to constrain the kinds of interpretations it is
possible to assign an expression used metaphorically. Indeed the greater
the role of its extralinguistic context in determining the content of a meta-
phor, the more we need to explain why only some and not other inter-
pretations can be expressed by the given expression. A primary function
of the notion of meaning is to furnish such constraints on interpretations
that draw on extralinguistic resources. So, rather than exclude a semantics
14 Chapter 1

of metaphor, its context-dependence exposes the proper role of semantic

knowledgeas knowledge of constraintsin metaphorical interpreta-
tion. Furthermore, although semantic competence by itself is not su-
cient for metaphorical interpretation and contextual input is absolutely
necessary, semantics plus context are all that are necessary. There is no
other special power or skill needed for a mastery of metaphor, even while
some of us are more accomplished masters of metaphorical use than
others.
This marriage between context-dependence and semantic knowledge
may appear to create an odd couple, but in fact metaphor is hardly
unique. There are many topics concerning the interaction between context
and languagefor example, speech-acts and conversationthat fall out-
side the scope of semantics, but one set of questions concerned with the
role of context in determining the interpretation of an expression on an
occasion is squarely semantic. Following David Kaplan (1989a, 522,
546), let's distinguish between the study of sentences-in-contexts and the
study of utterances and speech-acts. An utterance is a spatiotemporal
event that requires a speaker who uses a sentence to make a statement (or
perform some other act). An utterance is also subject to nonlinguistic
constraints, for example, the physical/biological fact that no human can
typically make more than one utterance at any one time and (if contexts
are individuated by their times) that no more than one utterance can
occur in any one context. A sentence, on the other hand, can occur in a
context even if the speaker of the context does not utter it; likewise, an
indenite number of sentences can simultaneously occur in one context (in
order to render it possible to evaluate them as the premises and conclu-
sion of a single argument). Given this distinction, we can then take the job
of semantics as the characterization of a sentence being true-in-a-context
rather than either the narrower study of sentences in isolation from all
context (or in the null context) or the much broader study of utterances
and speech acts. Semantics with this contextual edge is not new. The
analysis of a whole class of expressions of exactly this kindnamely, the
demonstratives (e.g., `this' and `that') and indexical expressions (e.g., `I',
`now', and `here')has long been assumed to fall within semantics. But
now, with David Kaplan's seminal work on demonstratives (a term I'll
use generically to include indexicals), we nally have the beginnings of a
rigorous formal theory that also does philosophical justice to the sub-
tleties of these context-dependent expressions.
To return now to metaphor: instead of allowing its context-dependence
to be an obstacle to its semantic candidacy, I'll argue that the key to its
Metaphorical Competence 15

satisfactory semantic analysis is to embrace its context-dependence. To go

one step further, I shall treat a metaphor as a type of context-dependent
expression of the same general kind as the demonstratives. What is
needed for such an account is to show that metaphors and demonstratives
share the same formal structure, to isolate the contextual parameter (like
the speaker for `I' and time for `now') that determines a metaphorical
interpretation, and to specify the rules that determine the contents of
metaphors in each of their contexts of utterance.22
To motivate the parallel between demonstratives and metaphors, I
might also mention that the same argument from context-dependence that
locates metaphors entirely outside semantics can be made for demonstra-
tivesand deserves the same kind of reply. Suppose (as Russell once did)
that a speaker's context-dependent knowledge of demonstrative reference
is simply knowledge of the entity to which one actually refers on each
occasion with a demonstrative token, like one's knowledge of the ref-
erent of a proper name. In that case, one's knowledge of demonstrative
reference would be entirely extralinguistic; there would be no isolable
language-specic knowledge the speaker would possess. What, if any-
thing, would be missing from such an extralinguistic account of demon-
strative reference?
Knowledge of the propositional content, or truth-conditions, of an
utterance containing a demonstrative, say, of `this is red', requires, of
course, extralinguistic knowledge of the thing demonstrated on the occa-
sion, the actual referent of the token of `this'. But when we understand this
utterance we also know the kind (however general it may be, such as an
object rather than property) of the extralinguistic entity that is admissible
as the referent of the demonstrative. We might, therefore, factor our
knowledge of demonstrative reference into two parts: (i) knowledge of the
kind (or range) of values that (any token of ) the demonstrative (type) can
be assigned on any occasion of use; and (ii) knowledge of the actual value
of the demonstrative (token) on a particular occasion of utterance. These
pieces of information are clearly distinguishable. A speaker-hearer can
know the rst without knowing the second, and a theory of the rst is
independent of the particular form we adopt for an account of the second.
But knowledge of the rst type is exactly what we would predict an object
of linguistic, or semantic, competence to be. Any account of demonstra-
tive reference that failed to distinguish it would fail to capture a signi-
cant dimension of the speaker's knowledge.
For the same reason, the speaker's knowledge of the type of parameter
on which a metaphorical interpretation depends should be distinguished
16 Chapter 1

from his knowledge of the value of the parameter in particular contexts

for particular utterances of metaphors. Knowledge of the rst kind, as we
would expect of linguistic knowledge, remains invariant from one to an-
other utterance of the metaphor despite the fact that the extralinguistic
factors that constitute the parameter will vary over contexts, yielding dif-
ferent interpretations for the metaphor on the dierent occasions. Here,
for metaphors as for demonstratives, the speaker's knowledge should
therefore be divided into two distinct components, (only) the rst of which
will fall in his semantic competence proper.
To articulate the dierence between these two types of knowledge
involved in metaphorical interpretation, I'll adopt David Kaplan's dis-
tinction for demonstratives between two semantic ``levels'' that were
originally conated in Frege's notion of sense. Kaplan calls the rst level
content, the second character. Content is (roughly) what we have been
calling the interpretation of a metaphor: what the metaphor says, its
propositional component, or truth-condition(al factor). So, just as the
content of a (singular) demonstrative is an object or individual, the con-
tent of a (predicative) metaphor is (something like) a property. Character
roughly corresponds to the (linguistic) meaning of an expression: a rule
known by speakers as part of their linguistic competence that determines
the content of the expression in each context of utterance (like the rule for
`I' that each of its utterances has its individual utterer as its content). Both
demonstratives and metaphors have nonconstant characters: characters
that determine dierent contents in dierent contexts.23 This is obvious in
the case of demonstratives but, as our earlier examples (1)(10) make
clear, the same is true for metaphor. One expression type, say, `is the sun',
is interpreted metaphorically in (1)(4), hence with one metaphorical
character; yet the one character yields dierent interpretations, or con-
tents, in dierent contexts (where, as I'll argue, the relevant dierence in
context is due to dierent contextual presuppositions).
An expression interpreted metaphorically has, then, a ``metaphorical
meaning'' in addition to a literal meaning. But this notion of meaning, it
should be noted, is rather dierent in kind from what previous writers
especially its detractorshave assumed a metaphorical meaning would
be. Its critics take the ``meaning'' of a metaphor to be something like the
property it expresses in a context. But that is what I take to be the content
of the metaphor in a context. On my account, the meaning of a metaphor
is the rule that determines its content for each context, that is, its charac-
ter. If we carefully bear in mind the character-content distinction, we can
Metaphorical Competence 17

answer a slew of questions that have standardly been raised to challenge

the legitimacy of a notion of metaphorical meaning and the possibility of
constructing a semantics for metaphor. For example:
1. If sentences (containing an expression) interpreted metaphorically are
semantically signicant, then, like other compound semantically valued
expressions, they must be compositional in structure. What kind of meta-
phorically relevant semantic structure can we articulate within them in
order that their truth-value be a function of the semantic values of their
parts?
2. It is a truism that the interpretation of a metaphor ``depends on its
literal meaning.'' Is that ``dependence'' semantic functionality? If so, what
is the function?
3. If the metaphorical interpretation of an expression F depends on its
literal meaning, then F must still ``have'' (in some sense) its literal mean-
ing, even while it is interpreted metaphorically. How? In what sense of
`have'?
4. If an utterance of a sentence interpreted metaphorically has a truth-
value that is (or might be) dierent from what its truth-value would be
were it interpreted literally, then we would also reasonably think that it
diers in meaning from its literal interpretation. What kind of meaning
might that be and how is it related to our notion of literal meaning? Does
metaphorical interpretation render an expression ambiguous or poly-
semous? Does the expression change its meaning when it is interpreted
metaphorically? How does this kind of change of meaning relate to other
changes of meaning?
Two additional consequences of the character-content distinction for
metaphor deserve mention. First, the ``received'' conception of semantics
``gives the meaning'' of sentences in the form of an account of the
speaker's knowledge of the full-blooded truth-conditions or propositional
contents directly assigned to the sentences. But semantics can maintain
this aim only as long as it innocently directs its attention at eternal, or
context-independent, expressions. When we focus on context-dependent
expressions like demonstratives, a direct assignment of truth-conditions to
expressions is ruled out. For these expressions, where context intervenes,
either we conclude that there is no semantic assignment at work, that their
``meaning'' lies outside the scope of semantics, or we must revise our
conception of semantics. Rather than concern itself with truth-conditions
or contents directly assigned to sentences, semantics now becomes a
theory of a speaker's knowledge of the form of truth-conditional or propo-
18 Chapter 1

sitional interpretations of sentences, that is, of the form or structure of

sentences that displays the contextual conditions or parameters with
respect to which their interpretations vary. In a word, semantics is now
the study of our knowledge of character rather than contentand simi-
larly for a semantics of metaphor.
Second, the nonconstant character of a demonstrative (e.g., the rule
for `I' that each of its tokens refers to its utterer) underdetermines its
content or interpretation, namely, an actual individual. For the same
character (e.g., of `I') yields dierent contents in dierent extralinguistic
contexts (namely, those containing dierent speakers). Hence any theory
of a speaker's semantic competence in demonstrativesa theory of her
knowledge of their characterwill be, in one sense, ``incomplete'' as an
account of her knowledge of their interpretation. Analogously for meta-
phors: Their interpretations, or contents, which also depend on all sorts of
extralinguistic abilities, beliefs, and associations, are underdetermined by
their respective characters, or meanings.24 As we saw when we contrasted
(1) rst uttered by Romeo in Shakespeare's context and then uttered by
Paris as a warning to Romeo, the same sentence, with the same meta-
phorical character, or meaning, can yield entirely dierent, even incom-
patible interpretations, given dierent presuppositions and attitudes. Thus a
semantic theory of metaphorical interpretationa theory of the speaker's
knowledge of the character of the metaphorwill also be ``incomplete''
as an account of her knowledge of the full interpretation. This kind of
incompleteness does not, however, signal a deciency in the approach.
Just the opposite. The virtue of the incompleteness is that it enables us to
discern the substantive contribution our semantic knowledge specic to
metaphor, our knowledge of its character, makes to our understanding of
metaphorical interpretation.
In sum, I shall argue that by exploiting the parallel between demon-
stratives and metaphors, we can identify a type of knowledge underlying a
speaker's ability to interpret a metaphor that belongs to his semantic
competence. Yet many readers may still not be persuaded that metaphor
should be explained by semantics, rather than by pragmatics or by a
theory of use. For the costs, one might object, especially the complica-
tions that accrue to our overall linguistic theory as a result of incorporat-
ing metaphor in semantics, outweigh the benets. In reply, I will argue,
the actual costs are negligible. We complicate our overall semantic theory
by including metaphor only if that requires us to introduce apparatus not
already and not independently necessary for the semantics. But if our
parallel between demonstratives and metaphors holds, the semantic rules
Metaphorical Competence 19

underlying our ability to interpret metaphors are of the very same kind as
those that underlie our ability to interpret demonstratives. Those rules (or
something like them) already constitute part of our semantic competence
in nonmetaphorical language, our competence in demonstratives. (All that
is additionally necessary, I will propose, is one general operator added to
the lexicon and one rule governing its operation.) So, if knowledge of
demonstratives belongs to linguistic competence, so should the cor-
responding knowledge governing metaphorical interpretation. Given a
semantics for demonstratives, metaphor can be had (virtually) for free.
With this overview of the argument in hand, let me conclude this sec-
tion with an outline of the chapters to follow.
Given my focus on context-dependence, the obvious, natural place to
locate metaphor within an all-inclusive linguistic theory is in pragmatics
or a use-oriented account; and this is, as we have said, the datum from
which many writers in fact conclude that metaphor should be explained
as a type of use or speech performance. To motivate my turn instead to
a semantic account of metaphor, I therefore begin, in chapter 2, with a
closer look at pragmatic or use theories. Concentrating on the inuential
essays of Donald Davidson, I argue that use theories need semantics pre-
cisely in order to constrain their too-powerful resourcesto explain why
specic expressions can be used to express only specic metaphorical
contents. A close critical look at Davidson's truth-theoretic semantic
treatment of context-dependence (e.g., demonstratives) within his use
theory also serves a second purpose: to motivate my own use of David
Kaplan's semantic framework that focuses on character rather than con-
tent for my account of metaphor.
With this motivation in hand, chapter 3 lays out the necessary semantic
background adopted from David Kaplan's seminal work on the logic of
demonstratives. I concentrate on two themes that play central roles in my
story: Kaplan's distinction between character and content and his inven-
tion of the operator `Dthat' to lexically represent demonstrative inter-
pretations (or uses) of (arbitrary, eternal) denite descriptions.
In chapter 4, I begin to lay out my semantic theory of metaphor as a
kind of context-dependent interpretation of an expression on the model of
demonstratives. I spell out the relevant feature of the context on which the
interpretation of a metaphor depends, a contextually given (sub)set of
presuppositions, and the semantic rule of character that constitutes the
meaning of the metaphor, the rule that determines its content or inter-
pretation in each context. In our earlier terminology, this is an account of
our semantic knowledge of metaphor.
20 Chapter 1

In chapter 5 I turn to the rst brand of our knowledge by metaphor,

concentrating on information conveyed as the content of the metaphor in
a particular context. This discussion helps my argument along in a num-
ber of ways, apart from oering a rst explanation of how our semantic
knowledge of metaphor enables us to express contents not expressible by
nonmetaphorical language. It also oers the rst detailed illustrations of
applications of our semantic knowledge to contextual presuppositions,
which in turn give a sharper view of the dierent kinds of knowledge
involved in the pragmatics of metaphorical interpretation and the not-
language-specic symbolic skills (such as the perception of similarity or
exemplication) that enter at this juncture. Finally, as part of my expla-
nation of these skills, I introduce the role of networks of expressions in
metaphorical interpretation; these networks also play an important role in
capturing the knowledge by metaphor conveyed at the level of meta-
phorical character, the topic of chapter 7.
Before turning to this last topic, however, I return in chapter 6 to the
relation between metaphorical character and meaning, and I use my
account to solve some outstanding problems about the pretheoretical
notion of metaphorical meaning and the formal problems raised by the
semantic data introduced at the end of chapter 2. Finally, I turn to the re-
lation of metaphor to other gurative and nonliteral uses of language, con-
trast my account with three other semantic theories in the literature, and
conclude the chapter with replies to a number of anticipated objections.
I nally turn, in chapter 7, to the ``character-istic'' information or sig-
nicance metaphors carry. This information is manifest in a variety of
waysin the explanatory power of beliefs containing metaphors, in the
sense of surprise associated with metaphor, and in the often repeated
claim that metaphors make us see one thing as another. Furthermore,
since this ``character-istic'' information is not expressed in the content of
the metaphor in a particular context, it is also not contained in para-
phrases of those contents. Through this variety of ways, our semantic
knowledge of a metaphorour knowledge of its characterenables us
to acquire information by the metaphor that we cannot grasp except
through knowledge of its character. On the one hand, then, the signi-
cance of a metaphor is not exhausted by its content in a context; on the
other, there is nothing about that signicance that is antithetical to a
semantics of metaphorical interpretation.
Finally, chapter 8 briey demarcates the boundaries of metaphor
linguistic versus pictorial metaphors, dead versus live metaphors, and the
literal versus the metaphorical.
Metaphorical Competence 21

III Methodological Preliminaries

Before beginning I want to raise several methodological and termino-

logical issues concerning (i) the unit of metaphorical interpretation, (ii)
the literal, (iii) the truth of metaphors, and (iv) the use of examples and
evidence. The rst three discussions will be brief, the fourth more detailed.

(i) The Unit of Metaphorical Interpretation

We should all agree both that the proper name `Juliet' has the same in-
terpretation (take your pick: extension, referent, content, intension, etc.)
in Romeo's utterance of
(1) Juliet is the sun
in Shakespeare's context as it has in
(11) Juliet is Romeo's beloved
and that `is the sun' (or `the sun') has a dierent interpretation in (1) than
it has in
(12) An especially bright star in our solar system is the sun.
We typically, and innocently, describe the dierent interpretation of `is
the sun' in (1) as ``metaphorical'' in contrast both to its interpretation in
(12), which we call ``literal,'' and to the interpretation of `Juliet' that
undergoes no change from (1) to (11). And based on this kind of familiar
data, we should say that the basic unit of metaphorical interpretation is
the specic subsentential constituent whose interpretation (whatever you
take that to be) undergoes change, namely, `is the sun', not the whole
sentence (1) and not even the pair of constituent expressions h`Juliet', `is
the sun'i.
Despite this seemingly straightforward way of identifying `the (a)'
metaphor, say, in (1), several philosophers and linguists have other can-
didates. Max Black (1993, 24) says that ``metaphor'' is always short for
``metaphorical statement'' and that ``statement-ingredients (words and
phrases used metaphorically)'' are only derivatively metaphorical.25
George Lako (1993) says that in ``contemporary research'' a metaphor is
really ``a cross-domain mapping in the conceptual system'' and that the
so-called metaphorical expression is nothing more than a linguistic item
that ``is a surface realization of such a cross-domain mapping'' (203).26
Eva Kittay (1987), who assumes a version of the deviance condition,
says that ``a unit of metaphor is any unit of discourse in which some
conceptual or conversational incongruity emerges'' (24), and within the
22 Chapter 1

metaphor she distinguishes a focus and frame. And, most recently, Roger
White (1996) has challenged ``the widespread assumption'' that ``in every
metaphor there is an isolable word or phrase which is the word or phrase
being used metaphorically'' (57).
What is at issue is not terminological but a matter of distinguishing
between the unit whose interpretation is being determined and the units
that determine the interpretation. Black is right to focus on the whole
statement insofar as its utterance is the minimal speech unit and all con-
stituents of an utterance can play a role in determining the metaphorical
interpretation of any single constituent.27 Lako is right insofar as it is
only generally as part of much larger linguistic (or, as he calls them, con-
ceptual) networks that individual expressions acquire their metaphorical
interpretations. Indeed I'll argue that the context in which an expression
is interpreted metaphorically must be broadened to include not only its
immediate linguistic environment, but also its extralinguistic situation
(including nonverbalized presuppositions and attitudes). But none of this
changes the fact that what is interpreted metaphorically in a context may
be a proper constituent within the sentence. Of course, the metaphorical
constituent is not always a simple expression rather than a phrase; more
than one expression can be interpreted metaphorically in a given utter-
ance; and we cannot always individuate or identify the metaphorical
constituent by looking merely at the (phonologically interpreted) surface
structure of the sentence uttered. It is also possible for an expression
interpreted metaphorically to be concurrently interpreted literally in the
same utterance in the same context, in which case it will be lexically am-
biguous, and it is possible for an utterance to admit multiple syntactic
analyses, each of which yields not only a metaphorical interpretation but
a dierent one. By saying that the metaphor can be a constituent expres-
sion, I should also not be taken to imply that its change in interpretation
(extension, referent, content, intension, etc.) ``exhausts its metaphorical
signicance.''28 As I'll argue at length in chapter 7, the character of a
metaphor carries information beyond that of its content (in its context),
part of which is also a function of the networks to which it belongs.
The question of the proper unit of metaphorical interpretation is bound
up with many issues we will take up in later chapters. For now, simply as
a matter of terminology, I shall mean by metaphor: `(a token of an) ex-
pression (type, simple or complex) that is interpreted metaphorically in its
context of utterance', and by sentence interpreted metaphorically or meta-
phorical sentence: `a sentence containing at least one expression that is
interpreted metaphorically in its context of utterance'. For brevity, I shall
Metaphorical Competence 23

also say that a metaphor is true or false, meaning that an utterance of the
(containing) sentence in which the given expression is interpreted meta-
phorically is true or false. Finally, a metaphorical sentence (statement,
utterance) is not something that is only metaphorically a sentence (etc.),
but a sentence (etc.) interpreted metaphorically.

(ii) The Literal

The term ``metaphor'' is often said to have two senses, one wide, one
narrow. In the wide sense, the metaphorical is contrasted with the literal
and includes the full range of nonliteral or gurative interpretations of
languageirony, metonymy, synecdoche, hyperbole, and so on. In the
narrow sense, the metaphorical is contrasted not only with the literal, but
also with these other gures or tropes. I shall use the expression (unless
noted otherwise) in the narrow sensemetaphor as distinguished from
both the literal and the other tropes. But (to anticipate my discussion in
ch. 6) I shall argue that our semantics for metaphor (in the narrow sense)
could be extended to certain of the other gures though not to others: to
metonymy and synecdoche, for example, but not to irony or hyperbole.
The notion of metaphor that will emerge from our analysis will be both
narrower and more variegated than the one in common circulation.
I rely on some distinction between the metaphorical and the literal and
I accept the truism (in a sense yet to be explicated) that the metaphorical
depends on the literal. But I shall not defend either of these assumptions
at this stage. For one thing, critics who baldly deny the distinction owe us,
in my view, a clear statement of what they think they are denying; for
another, I cannot yet clearly articulate the distinction I wish to draw
without much more groundwork. Indeed, dierent notions of the literal
will emerge in the coming chapters and, as I'll argue in chapter 8, the
notion of the literal is in worse theoretical shape than the metaphorical.
As a working hypothesis, I shall assume that the literal meaning of a
simple expression is whatever, according to our best linguistic theory,
turns out to be its semantic interpretation, and that the literal meaning of
a sentence is the rule-by-rule composition of the literal meanings of its
simple constituents.29 As a matter of fact, current semantic theory is not
yet in a position to state with any authority what the semantic interpre-
tation of a simple expression is, but it should also be noted that we do
know it is nothing like a set of necessary and sucient descriptive con-
ditions. Minimally, it contains the extension or referent of the expression
and the constraints and conditions that govern both its interaction with
the syntax and with the extralinguistic context.
24 Chapter 1

Finally, to explain how the metaphorical interpretation of a token of (a

type) F depends on its literal meaning, I'll need the vocabulary to say that
the token (in one sense) ``has'' its literal meaning and (in a second sense)
does not ``have'' it. For that purpose, I'll use the term of art literal vehicle
to refer to the expression type F with its literal meaning when the token of
F is interpreted, or used, metaphorically.

(iii) Metaphors and Truth

In proposing a semantic theory of metaphor, I have assumed (without
argument) that metaphorsor declarative sentences in which at least one
expression is a metaphorare truth-bearing entities; that they are true or
false no dierent from literal sentences like `snow is white'; equivalently,
that they express propositions. This assumption is far from uncontro-
versial, so it might be objected that I am assuming the crux of what I
claim to demonstrate in this book.
For a start, the best defense is a good oense. It certainly looks like
metaphorical (declarative) sentences are true or false. Many of our inno-
cent assertions employ metaphors. To the man on the street unversed in
philosophical semantics, the assumption that utterances containing meta-
phors are true or false (as the case may be) would be beyond reproach.
When Romeo utters (1), he not only wants to ``call our attention'' to a
(particular) similarity between Juliet and the sun; he also intends to say
something true about Juliet, to assert that she has a certain property (or
set of properties) ``corresponding'' to the predicate `is the sun'. Now,
whatever it is that he is asserting, and thereby representing himself as
believing to be true, it is not what is expressed by (1) interpreted literally.
What, then, would be simpler than claiming that the property he believes
to be true of Juliet is a property metaphorically expressed by the predicate
`is the sun'? That he is asserting a proposition metaphorically expressed
by (1) that he believes to be true?
Suppose that Count Paris disagrees with Romeo's utterance of (1). He
is surely not denying the proposition expressed by (1) interpreted literally.
Romeo and Paris agree about that proposition that it is false. So what is
the common thing Romeo asserts and Paris denies? Isn't it the proposition
asserted by (1) when it is interpreted metaphorically? Indeed why not?
In short, the ordinary appearance is that utterances of sentences that
contain metaphors are truth-valued, express propositions, and can be
used to make assertions (or other speech acts that presuppose assertion).
The burden of argument, therefore, falls on those who deny that this
appearance is reality. To be sure, there is no lack of arguments for the
Metaphorical Competence 25

other side. These range from observations based on our ordinary use of
metaphors (to ``call our attention to a certain likeness,'' or ``invite'' us to
``appreciate'' a resemblance, or ``inspire'' a certain vision, or ``propose''
that things be viewed a certain way) to theoretical considerations about
compositionality and the formal structure of a semantic theory. I shall
address these objections in the course of the book, but I defer them until I
can rst set out my own alternative theory.
One last methodological remark on this issue: Despite the ordinary
presumption to which I have appealed, the thesis that metaphorical in-
terpretation falls within the scope of semantics cannot be settled simply by
appeal to ``facts'' like our practices to use metaphors in assertions. On the
one hand, actual practice can always be interpreted and explained in a
variety of ways consistent both with the assumption that metaphors are
truth-valued and with the assumption that they aren't. On the other hand,
even if ordinary practice were dierent, the decision to treat metaphors as
truth-bearers could be justied on theoretical grounds. Truth-values are
theoretical entities. They serve as the semantic values or roles of sentences
in a complex, systematic, powerful theoretical framework that aims to
account for our understanding of language. If this same framework pro-
vides an illuminating account of metaphor, the assumption that meta-
phors are truth-bearers will be warrantedlike any theoretical posit that
is justied by the evidence for its containing theory and by its explanatory
success.

(iv) Examples and Evidence: Living vs. Dead Metaphors

Theories of metaphor are often a function of their authors' examples.30
Philosophers and linguists focus on metaphors heard in ``ordinary
speech''; no wonder, it is charged, that their theories best, or only, t
``conventional,'' ``frozen,'' or ``dead'' metaphors. Literary critics and
rhetoricians analyze the ``novel,'' ``imaginative,'' and ``creative'' meta-
phors of poetry and literaturewhose complexity and subtlety tend to
make them suspicious of the possibility of any linguistic theory of meta-
phor, period. As a description of current practice, this observation con-
tains a grain of truth. However, some theorists go further, claiming that
there are also good reasons to take one or the other kind of example as the
paradigm of a metaphor. Others, typically in the course of polemically
defending their own theory, charge that counterexamples of the other kind
are not ``real'' metaphorshence are not counterexamples.31 A familiar
complaint of this sort is that philosophers' and linguists' theories are in-
adequate to deal with the subtleties of poetic metaphors because they are
26 Chapter 1

based on simple (or is it simplistic?), tired, and sometimes dead examples

(e.g., `Man is a wolf ').32 On the other hand, certain linguists have recently
claimed that theories that focus on poetic metaphors miss the system-
aticity and conventionality exemplied by the ubiquitous metaphors of
ordinary, normal speech, properties that these authors further argue are
essential characteristics of metaphor (and, once identied in ordinary
speech, can also be found to underlie poetic metaphors).33
These debates frequently appeal to the distinction between live (or,
better, living) and dead metaphors. This is a time-worn distinction calling
out for reconsideration, to which I shall return in chapter 8. Here I want
to address a prior, methodological question: Are there principles that
ought to govern our selection of examples? Are some types of metaphors
to be preferred over others? What of the charge that a theory ignores a
whole sample space of metaphors?
To begin with, I want to distance myself both from the populism
advocated by the spokesmen for metaphors of ordinary speech and from
the elitism fostered by the cognoscenti of poetry. From a semantic point
of view, there is one metaphorical competence that underlies our ability to
produce and comprehend all metaphors regardless of their context of use,
be it poetry or ordinary speech. All that distinguishes these dierent met-
aphors are the dierent skills and sensibilities recruited in addition to our
semantic competence to complete their actual interpretations in their re-
spective contexts. At the level of semantic explanation, the linguistic phe-
nomenon of metaphor should not be identied exclusively, or even too
closely, with any one particular brand, or use, of metaphors.
On the other hand, the evaluation of a semantic theory of metaphor
need not depend, rst and foremost, on whether it ``covers all the data''
equally well. As evidence for or against a particular theory, it does not
follow that one metaphor is never more relevant than another. Not that
there are ``don't cares.'' Every theory must account for some classsome
interesting classof data, but no theory must (or can) account for ``all''
the data. A theory need only account for the relevant data and, most im-
portant, it is the theory itself that provides the criterion of relevance.34
Examples, to be examples, should be representative of the explanandum,
but what counts as representative of a phenomenon X cannot be deter-
mined independently of, or prior to, a particular theory of X.
These general lessons apply equally to metaphor. Our concern is with
the semantic competence underlying metaphorical interpretation that
consists in knowledge of context-sensitive rules that determine the struc-
ture of metaphorical interpretations; it is not with the interpretations
Metaphorical Competence 27

themselves or with their eects. Our examples should accordingly be ones

that bear on the context-dependent structure of those rules and, not sur-
prisingly, these will be metaphors whose interpretations maximally de-
pend on their context``living'' metaphors in one sense of the term.
Interpretations that are dead to their context of utterancethat are called
``metaphorical'' only because of their historical originsimply do not tell
us much, pro or con, about the claims of our theory. And the more bor-
derlinethe less context-sensitivethe example, the less desirable it will
therefore be for our purposes. But matters could be otherwise. For an-
other purpose, say, to explain why some but not other metaphors endure,
or to explain what makes a metaphor rhetorically or aesthetically eec-
tive, it may well turn out that (certain) dead metaphors are better exam-
ples than live ones. As Borges reminds us:
When I was a young man I was always hunting for new metaphors. Then I found
out that really good metaphors are always the same. I mean you compare time to
a road, death to sleeping, life to dreaming, and those are the great metaphors in
literature because they correspond to something essential. If you invent meta-
phors, they are apt to be surprising during the fraction of a second, but they strike
no deep emotion.35
Apart from metaphors whose interpretations are relatively dead to their
contexts, I should also mention here a second class of metaphors that are
not maximally germane to my theory. These are metaphorical interpreta-
tions that extend the (literal) meaning of an expression, say, by dropping
at least one condition in the (literal) meaning, in contrast to interpreta-
tions that involve changes (say, of extension) of the type Aristotle called
transfer. When we say, for example, that Quine demolished Carnap's argu-
ment, we drop, as it were, one or another condition of application asso-
ciated with the (italicized) term under its literal interpretation, so that
the resulting metaphorical interpretation extends, and properly contains,
the original one. Such extensions might be context-dependent in that the
context determines the conditions to be dropped on the occasion, and
dierent conditions might be dropped in dierent contexts. However,
these extended interpretations are context-independent insofar as they do
not draw upon extralinguistic presuppositions for the content of the
metaphorical interpretation.36 In the case of transfer, we interpret the
expression to express a particular content depending on contextual pre-
suppositions somehow associated with the term, such that the resulting
interpretation is applicable to a domain disjoint from its original one (or
suciently disjoint for the change to count as one of transfer). For ex-
ample, what is metaphorically expressed by `the sun' in `Juliet is the sun'
28 Chapter 1

is transferred from, rather than an extension of, its literal meaning. For
the purposes of my theory, we will be primarily concerned, then, with
transferred rather than merely extended interpretationseven though my
account can be broadened to cover extended interpretations.
In sum, a theory (like mine) that claims to apply to all metaphorical
interpretations need not, and typically will not, be equally conrmed by
all metaphors (or disconrmed by just any). Furthermore, which meta-
phors are germane to the evaluation of the theory will be determined, at
least in part, by the very theory. Finally, the kind of metaphor that is
germane evidence for my semantic theory, I now want to argue, is not
appropriately described either as the conventional, dead metaphor of
ordinary speech or as the novel, creative, living metaphor of poetry.
Let me return for a minute to the distinction between dead and living
metaphors. Although I will defer a full discussion of what makes a meta-
phorical interpretation living or dead until chapter 8, I have already sug-
gested that the liveliness (in one sense) of a metaphor is at least in part a
function of its degree of dependence on its context. Notice, however, that
this distinction between the living and the dead is not between kinds of
expressions but between interpretations in contexts. Obviously some meta-
phorical interpretations of some expressions are dead in some contexts,
but even the received interpretation of a time-honored dead metaphor like
`leg of a chair' only happens to be dead most of the time in most contexts.
The same expression might yet be given a dierent living metaphorical
interpretation; indeed even its dead metaphorical interpretation might be
brought back to life or resuscitated in another context. Someone might
tell us to look at the sexy legs of a couch; `hot as hell' gets new life as `hot
as the hinges of hell', and `full of wind' becomes, in a poem of Yeats, `an
old bellows full of angry wind'. Sometimes, too, we can verbally resusci-
tate a dead metaphor by extending it, that is, by making explicit the
family of metaphors to which it belongs. Thus each metaphor in the fol-
lowing passage would be more or less dead standing alone in an isolated
context. But juxtapose them and they start breathing with life:
Although George's own claims were indefensible, he attacked every weak point in
my argument. I won the argument with himdespite his criticisms which were
often right on target and despite his attempts to shoot down all my own claims
only because I managed to demolish him.37

Furthermore, whether a metaphor is dead or living can never be deter-

mined simply by ``looking'' at it, either at the concrete expression itself or
at its metaphorical interpretation.38 We must also know how that meta-
Metaphorical Competence 29

phorical interpretation was arrived at in its respective context, which de-

pends both on its logical form (or character) and on the role of the context
in the assignment. And this way of distinguishing living and dead meta-
phors cuts across the distinction between poetry and ordinary speech.
Living, creative, or novel metaphorical interpretations can be found in
ordinary exchanges as well as in poetry or literature, and dead metaphors
might equally well turn up, and be used eectively, in poetry (or literature,
as Borges said). Time Magazine, one of my own favorite sources of meta-
phors, is a publication whose understanding surely requires no special liter-
ary sensibility; yet its metaphors are frequently novel, imaginative, witty,
and full of life. Here is one example:
At Checkpoint Charlie, the hideous maw of the Berlin Wall gapes briey, aord-
ing a narrow passage into the divided German soul. On its Western side, a sea of
sensuous color rushes down the Kurdamm, past the ruins of the Kaiser Wilhelm
Memorial Church, and spends itself violently but impotently in a scatological orgy
of grati against the cold barrier. . . . Propelled by the engine of the postwar
Wirtschaftswunder, the capitalist Federal Republic of Germany is a sporty blond
racing along the autobahns in a glittering Mercedes-Benz. The Communist Ger-
man Democratic Republic, bumping down potholed roads in proletarian Wart-
burgs and Russian-built Ladas, is her homely sister, a war bride locked in a
loveless marriage with a former neighbor. (Time, March 25, 1985)
In sum, living metaphors should not be identied with poetic metaphors.
Although we have good reason to prefer examples that are living (i.e.,
maximally context-sensitive) metaphors, we need not look specically to
poetry to nd them.
Nonetheless, despite the reasons I have givenand indeed, one sus-
pects, despite any reasons anyone might givemany, especially literary,
theorists will be unmoved by what I have said. They will insist that we
misrepresent the nature of metaphor if we really think that ``ordinary''
metaphors are as vital and as representative of the phenomenon as poetic
metaphors. What more can we respond to this point of view?
Resistance of this kind runs very deep, resting on deeply ingrained
attitudes rather than on reasoned arguments, on attitudes for which there
are undoubtedly a number of sources, however dicult it is to pinpoint
them.39 One source might be Aristotle's well-known claim that ``true''
metaphors are a ``sign of genius'' (Poetics 1458b), though he does not
single out poetic metaphors in this passage. Yet, insofar as the ordinary
speaker's understanding of his metaphors does not involve genius, it
might be thought that Aristotle is describing one special class of meta-
phors, namely, those of poetry.40
30 Chapter 1

A second source is a complex of motifs. One of these is the identication

of metaphor and poetry. Each live, novel, or creative metaphor is treated
as if it were itself a work of art or poetry; alternatively, the most basic
form of literary art, or poetry, is claimed to be the individual metaphor.41
A second motif holds that there is a subject-specic ``proper'' mode of
inquiry to study, or understand, works of art or poetry, and hence, to
study or understand metaphor. It is almost impossible to spell out this
``method'' but the idea rests on an opposition between the humanistic
disciplines, exemplied by poetry and art, and the methods of physical
science. Underlying this second motif we can already detect a slew of
familiar dichotomies: science versus art, reason versus feeling, poetic ver-
sus nonpoetic language.42 A third motif is that metaphor is claimed to be
the essential kind of language in general, ``the most vital principle of lan-
guage and perhaps of all symbolism.''43 In this last step, the attitude that
at rst was specic to metaphor is now generalized to embrace all lan-
guage: It opposes any attempt to subject language to naturalistic methods
of study and instead views it as if it were ``something `higher,' mysterious,
`spiritual,' '' something that cannot be studied on a par with natural
phenomena.44
Like many attitudes, this stance toward metaphor is not one against
which one can argue directly. We can change peoples' attitudes of this
sort only by addressing their underlying fears and concerns, by showing
them how to do what they fear cannot be done by a theory. We must
subject metaphor to rigorous analysis and prove by example that to do so
is not to do injustice to its subtleties.
One nal methodological point about the opposition between the poetic
and nonpoetic: Those who insist on this dichotomy only emphasize the
dierence between metaphors in and out of poetry; they indiscriminately
lump together all (living) metaphors within poetry without acknowledging
their signicant dierences. Among living poetic metaphors, I would also
argue, there are some our theory should not attempt to explain, at least
not directly.
To borrow some terminology from Chomsky, let's distinguish core as
opposed to peripheral metaphors. This is not a distinction of kind but of
degree and it is highly theory relative. If the aim of our semantic theory of
metaphor is to discover the general (and presumably universal) semantic
principles governing metaphorical interpretation, core metaphors are
those whose properties most directly bear on those principles; peripheral
metaphors are those whose exceptional properties require signicant
additions to our general theory. There is also a second way to draw this
Metaphorical Competence 31

distinction: To the extent to which the general principles of metaphorical

interpretation are semantic universals (or are determined by such univer-
sals), core metaphors whose properties are explained by those semantic
principles involve no true learning. Peripheral metaphors, on the other
hand, are marked; to the degree to which they involve exceptional prop-
erties, that is, properties not explained by universal semantic principles,
their rules of interpretation must be explicitly learned. Both core and
peripheral metaphors may, then, be dependent on context-specic pre-
suppositions (the content of which, if extralinguistic, must be learned for
both) and be dependent in one, specically metaphorical way. But they
dier in that peripheral metaphors require, in addition, training or learn-
ing in order for one to fully grasp their technique of interpretation, a
technique that still presupposes the semantic competence to interpret
metaphors.45 Now, there are very specic, dierent kinds of poetic meta-
phors that are peripheral in this sense, but they are highly prominent in
the work of modernist, surrealist, symbolist poets, such as Rilke, Celan,
Mayikovsky, or Pound. These metaphors require a theory of interpreta-
tion, but it will not consist solely of our semantic theory (supplemented by
our knowledge of the extralinguistic context).46 By distinguishing them
from the core metaphors, my point is not that we should disregard them.
Rather, if we give them undue emphasis, we obscure the general principles
of metaphorical competence they presuppose. And it is these general
principles that are our primary focus in this book.
Chapter 2
From Metaphorical Use to
Metaphorical Meaning

Few distinctions are more basic to twentieth-century philosophy of lan-

guage than that between ``what words mean and what they are used to
do.''1 And if a philosopher of language trained in this century were asked
to explain the phenomenon of metaphor, your best bet is that his gut re-
action would be to treat it as a use of language rather than as a type of
meaning, by a pragmatic rather than semantic theory. Now, a theory
of use is the natural starting place for an explanation of metaphorand
especially if one focuses, as I do, on its context-dependence. My aim in
this chapter is to show why we nonetheless need a semantics of metaphor.
That is, I shall argue for a theory of metaphorical meaning by demon-
strating, given the assumption that use in a context is essential to their
interpretation, that it is necessary to posit meanings for metaphors in
order to constrain their use.
There are any number of use, pragmatic, speech-act, or conversational
theories of metaphor in circulation, but among them we can distinguish
two camps. The rst holds that just as there are semantic theories, so there
are pragmatic theories, including pragmatic theories of metaphor. Some
of these theories treat metaphor as an instance of speaker's meaning like
an indirect speech act (Searle 1993), some as a type of gurative speech
act (Cohen 1975; Loewenberg 1975), others as a brand of calculated
conversational implicature (Grice 1975; Sperber and Wilson 1985/6), and
yet others as a form of Gricean interpretation of speech that depends on
mutual recognition of intentions (Fogelin 1994). The second camp takes
metaphor to be a matter of use in the Wittgensteinian sense, according
to which it eludes explanation in terms of regularities or rules, includ-
ing those captured by notions like speaker's meaning, illocutions, per-
locutions, or (violations of ) maxims of conversation. The best-known
representative of this second camp, Donald Davidson, denies both the
34 Chapter 2

possibility of a theory of metaphorby which he means a nitistic rule-

like account of metaphorical interpretationand of any nonvacuous
explanatory notion of metaphorical meaning.2 What a metaphor conveys
is largely nonpropositional, the product of a non-rule-governed creative
skill, an unpredictable causal eect of its utterance-event understood
exclusively according to the literal meaning of its words.
The dierence between these two approaches to metaphor reects a
deeper division over the status of pragmatics or use. For philosophers like
Searle semantics and pragmatics are each theories, with their own respec-
tive principles of meaning, corresponding to dierent domains of phe-
nomena. For Davidson the very idea of pragmatics as a theory of a
domain marked o as use of language is a misnomer. Instead, as he sees
it, in the beginning was the utterance, which we identify as intentional
and linguistic and go on to interpret. Where we can identify regularities
across utterances, and across our interpretive practices, we abstract, with
the usual degree of idealization, a theory of meaning or semantics. To
the extent to which utterances, or language use, submit to any sort of
theory of interpretation, that theory is semantics. Language use like meta-
phor that does not submit to semantic theory does not submit to another
kind of theory; instead it is use of language that resists all theoretical ex-
planation of the type provided by principles of meaning. If it can be
``explained,'' it will only be causally.
I'll begin this chapter with a quick look at two theories representative
of the rst camp, Grice's and Searle's. (I'll return to other pragmatic
accounts, such Fogelin's, in later chapters.) However, I shall mainly focus
on Davidson, whose inuential paper articulates a widespread attitude
among philosophers and poses the most serious challenge in the literature
to my position.
Davidson explicitly makes two negative claims ((1) and (2)), the rst of
which comes in two versions, and one positive proposal (3). In addition,
his account makes an implied third negative claim (4):
(1a) Individual words in a metaphorical utterance do not have a
``metaphorical meaning'' in addition to or in place of their literal
meaning; instead they exclusively retain their literal meaning.
(1b) The sentence used in a metaphorical utterance does not have a
``metaphorical meaning'' in addition to or in place of its literal meaning;
instead it exclusively retains its literal meaning.
(2) Apart from the errors of ``meaning-ication'' in (1a,b), a metaphor
does not have ``a denite cognitive content that its author wishes to
From Metaphorical Use to Metaphorical Meaning 35

convey and that the interpreter must grasp if he is to get the message''
(WMM 262).
(3) A metaphor is an imaginative use of a sentence, exclusively with its
ordinary literal meaning, whose intended eect is to make us notice a
likeness.
(4) There can be no (nitistic) theory that would show how the
metaphorical meaning (or metaphorical propositional content), if there
were one, of every metaphor expressible in a language is a function of a
nite number of features (or a nite number of meaning-axioms for a
nite number of simple expressions in the language) and a nite number
of rules of composition. That is: there can be no semantic theory of
metaphor.
In this chapter I shall bracket Davidson's arguments for (2) and return
to them in chapter 7 when I take up ``knowledge by metaphor.'' The
fourth claim, Davidson's denial of the possibility of a semantic theory of
metaphor, is the thesis most directly opposed to my position, but he does
not explicitly defend it. I'll return to some possible considerations in its
support in the last section of this chapter. However, a satisfying (or, at
least, satisfactory) reply will come only when we sit down together to eat
the pudding I shall begin concocting in the next chapter. What Davidson
states in WMM are necessary conditions that any notion of metaphorical
meaninghence any notion to be explained by a semantic theory of meta-
phormust satisfy, and, throughout the essay, he directs us to a number
of prima facie obstacles that stand in the way of straightforwardly meet-
ing those conditions for metaphor. In this chapter I shall concentrate on
these conditions for meaning and Davidson's reasons for thinking that
purported metaphorical meanings cannot satisfy them. At the same time,
I shall assess his complementary proposal that ``how metaphor works''
can be adequately explained in terms of the use of sentences with nothing
but their literal meaning. By pinpointing the inadequacies in Davidson's
story, I hope to show why a use ``theory'' needs a notion of metaphorical
meaning. Finally, I shall try to show why such a notion of meaning can-
not be adequately captured in Davidson's own general theory of meaning
(as a theory of truth-conditions). The problem, I shall argue, reects a
more general diculty in Davidson's treatment of context-dependent ex-
pressions, and this problem will motivate my shift (in ch. 3) to Kaplan's
conception of semantic theory based on the character-content distinction.3
In WMM Davidson explicitly presents both his critique and positive
proposal only in a few lean sentences. But his argument can be recon-
36 Chapter 2

structed in light of his other writings on radical interpretation and, in

particular, one of his most recent papers, ``A Nice Derangement of Epi-
taphs.''4 Although my aim is not Davidsonian exegesis, I will freely weave
among these papers to construct as strong a case as I can.

I Meaning vs. Use

According to Davidson, ``metaphors mean what the words, in their most

literal interpretation, mean and nothing more'' (WMM 246, my italics).
``Nothing more'' excludes ``semantic resources beyond . . . the ordi-
nary'' (WMM 245, my italics), special metaphorical word- or sentence-
meanings, and also anything a metaphor might be thought to convey as
utterance- or speaker's meaning a la Grice and Searle. As Davidson adds
parenthetically: ``nor does [the] maker [of a metaphor] say anything, in
using the metaphor, beyond the literal'' (WMM 247). Davidson's remark
is not specic to his view of metaphor. He is skeptical in general about the
possibility of codifying the abilities and skills involved in so-called
speaker's or utterance-meaning in the form of ``principles'' (a la Searle) or
``maxims'' (a la Grice); the kinds of inferences and reasoning these activ-
ities employ require only the ``cleverness,'' ``intuition, luck, and skill'' that
are necessary for any rational activity or for ``devising a new theory in
any eld.''5 What words are used to do should never be considered
meaning, even when qualied by the terms ``speaker'' or ``utterance.''
Here is not the place for a full evaluation of conversational implicatures
and speaker's meaning as candidates for meaning. However, I agree with
Davidson that metaphor, insofar as it is a matter of use, should not
be subsumed under speaker's or utterance-meaning. A quick glance at
Grice's and Searle's treatments should make clear why.6
Grice takes up metaphorfor example, `you are the cream in my
coee'in order to illustrate how the outing of a maxim (e.g., Quality)
leads to a conversational implicature. The reasoning underlying the
implicature consists in two steps. First, because metaphors ``characteris-
tically involve categorial falsity'' when they are taken literally (according
to ``what the speaker has made as if to say''), we infer that the speaker
could not have meant that. But neither could he have meant its contra-
dictory (as in irony) because that interpretation in turn would be a truism.
Therefore, at the second step, we infer as ``the most likely supposition . . .
that the speaker is attributing to his audience some feature or features in
respect of which the audience resembles (more or less fancifully) the
mentioned substance.''7
From Metaphorical Use to Metaphorical Meaning 37

Notice that the maxim of Quality functions in this calculation to con-

versationally implicate only that the utterance is not to be taken literally
(or, for that matter, ironically) and that it is to be interpreted metaphori-
cally.8 That is, the stage of Grice's explanation that actually employs a
violation of a conversational maxim only bears on our knowledge that the
utterance is a metaphor, the recognition task, not the task of interpreta-
tion, our determination of what the metaphor means. When Grice turns
to the latter task, he appeals to a probable (``the most likely'') supposition
involving resemblance, that is, an ad hoc though traditional principle
added to the conversational maxims that involves no calculation or
implicature. Hence even if we were to count calculated implicatures that
result from outings of maxims as ``meanings,'' the content of the meta-
phorical interpretation would still not be such a product. On Grice's own
grounds, we have been given no reason to consider the feature of resem-
blance that constitutes the content of the metaphor as a kind of implica-
tionally derived meaning.9
Searle, too, counts metaphorical interpretation as a kind of ``speaker's
utterance meaning'' as opposed to sentence meaning. He begins from the
premise that our ability to understand a metaphor, although not part of
our semantic competence (which governs sentence meaning), is nonethe-
less ``systematic rather than random or ad hoc.'' Hence, there must be
``principles'' that explain how speakers ``can say metaphorically `S is P'
and mean `S is R,' where P plainly does not mean R'' (Searle 1993, 113).
Searle emphasizes that there is no single principle that explains how this is
achieved for all metaphors; instead he proposes eight dierent principles,
adding the ``condent'' proviso that they are not exhaustive. However, all
the principles, he adds, ``call R to mind'' given an utterance of P.
Let's grant for now that the various principles indeed call R to mind.
Does the eectiveness of the ``principles'' justify Searle's claim that R is
the meaning of the metaphorical use of P?
Two points emerge when we examine Searle's eight principles. First,
they demonstrate that the contents or features R that serve as metaphori-
cal interpretations of expressions P are as heterogeneous as can be. The
Rs can range over features that are either denitionally (Principle 1) or
accidentally (Principle 2) true of the Ps but in general they need not be
true or even be believed to be true of the Ps. An R feature may only be
culturally or naturally ``associated with P in our minds'' (Principle 4), or
R may simply hold under a condition somehow ``like'' that of P (Principle
5). There is, in other words, no single principle (e.g., resemblance) that
would describe the contents of all metaphorical interpretations; the best
38 Chapter 2

we can do is say that P ``calls [R] to mind,'' describing the range of Rs

case by case. For this last descriptive function, Searle's eight principles
provide us with a helpful, if rough, catalog of what can serve as an inter-
pretation of a metaphorically used expression, the range of possible values
of R for dierent Ps.
But this descriptive strength of Searle's principles is also their explana-
tory weakness. Because of their descriptive ecumenicalism, the principles
place no restrictions on the class of possible features that can enter into a
given metaphorical interpretation. What one principle rules out, another
rules back in. Furthermore, it is completely obscure what it is for one
thing to call another to mind. No one knows what kind of psychological
ability or complex of abilities this is; at the very least, it is no better un-
derstood than the phenomenon of metaphor it is meant to explain. Searle
himself comes close to acknowledging the explanatory poverty of his ac-
count when he says that the ``question, `How do metaphors work?' is a bit
like the question `How does one thing remind us of another thing?' ''
However, he thinks that the two questions are alike because ``there is no
single answer to either'' (my italics). One might rather say, as philoso-
phers since Hume have recognized, that what they have in common is that
we know no answer to either of them. More important, even with an ex-
planation of this psychological phenomenon, it should be obvious that
not everything that something calls to mind, or reminds us of, is some-
thing it means. That a metaphor calls a feature R to mind is hardly su-
cient to count R as part of its meaning. What would be necessary for
something to count as metaphorical meaning is not only that it achieves
an end but that it does so by providing a distinctive means``the meta-
phorical way of using or interpreting words,'' whatever that turns out to
be. ``Reminding'' or ``calling to mind'' simpliciter can't do that job. In
sum, Searle's eight principles, however valuable descriptively, provide no
explanation of how P imparts R or why such a means should count as
meaning.
If these brief remarks are on the right track, then, as Davidson also
thinks, we gain little explanatory value by thinking of the features com-
municated by a metaphor as its utterance- or speaker's meaning. Better
not to use ``meaning'' when all we mean is use. However, Davidson's own
thin remarks criticizing metaphorical meaning suggest another direction
to explore in search of a candidate. What is wrong in ``posit[ing] meta-
phorical or gurative meanings, or special kinds of poetic or metaphorical
truth,'' in order to explain how ``words work in metaphor'' is, Davidson
says, that this is
From Metaphorical Use to Metaphorical Meaning 39

like explaining why a pill puts you to sleep by saying it has a dormative power.
Literal meaning and literal truth conditions can be assigned to words and sen-
tences apart from particular contexts of use. This is why adverting to them has
genuine explanatory power. (WMM 247)

Here Davidson says that appeal to metaphorical meaning is explanatorily

vacuous for a reason connected somehow to its context-dependence. Thus
a feature like one of Searle's Rs can't be a metaphorical ``meaning''
because (i) it ``is not a feature of the word that the word has prior to
and independent of the context of use,'' and (ii) unlike literal meaning
that explains why all utterances of one sentence have the same truth-
conditions, there are no analogous cross-contextual regularities to explain
for metaphor: Each metaphorical utterance in its context appears to ex-
press a dierent feature from every other one.10 Now, this criticism can
be turned on its head: If we could nd some other candidate (dierent
from one of Searle's features R) that did meet these conditions, we would
be well on our way toward a notion of metaphorical meaning. Rather
than refuting the possibility of metaphorical meaning, we can read
Davidson as stipulating two adequacy-conditions to be met by a kosher
notion of metaphorical meaning. In the coming sections we shall try to get
a better grip both on these conditions and on the kind of candidate for
metaphorical meaning that might satisfy them.

II If Literal Meaning, Why Not Metaphorical Meaning?

To understand better why a (metaphorical) meaning must be ``prior to

and independent'' of its context of use, let me turn rst to Davidson's
conception of literal meaning and its place in a theory of use.11
Following Davidson, a theory of language is at bottom a theory of
linguistic use, that is, a theory of utterances to be explained as a species of
rational acts ultimately performed for the purpose of communication.12
Communication succeeds when a speaker S's utterance is interpreted as
he intends, that is, when the hearer or, as Davidson prefers, interpreter I
understands S's utterance just as S intended it to be understood. In that
case, Davidson argues, we can shift the explicit object of our theorizing
away from S to I 's knowledge that enables him to interpret S. Further-
more, since we often communicate with utterances of ungrammatical as
well as grammatical strings, Davidson assumes that a theory of interpre-
tation must account for them all, including (as we discover in NDE)
malapropisms, slips of the tongue, and half-nished sentence fragments.
Finally, because communication occurs when I understands what S
40 Chapter 2

intends to say, a theory of interpretation should be ``adequate'' (NDE

444) to the requirements of shared (linguistic) understanding.
It is this focus on understanding that yields Davidson's well-known
proposal that a theory of interpretation takes the form of a Tarskian
theory of truth. However, it is helpful to distinguish two distinct routes by
which understanding leads to this conclusion. The rst route attempts to
negotiate its way between one claim that given the holistic nature of (lin-
guistic) understanding, any theory of interpretation of a single utterance
must provide interpretations for all of its speaker's potentially innite
number of novel utterances, and a second claim that the theory must
represent this interpretive ability in a form that respects our nite human
capacities. The theory accomplishes this by showing how the interpreta-
tion of each utterance is a function of the interpretations of its compo-
nents: elements drawn from a nite stock of basic vocabulary composed
into more complex interpretations by a nite stock of rules of composi-
tion. At this point Davidson brings on stage Tarski's theory of truth as an
``adequate'' theory of interpretation with exactly this formal structure.
But technical matters aside, there is also a second route by which under-
standing leads to the same destination. Recall that the fundamental no-
tion in Davidson's communication-oriented theory of interpretation is
shared understanding, and it is joint knowledge of the truth-conditions
of the utterance that S and I share when communication succeeds at
achieving this kind of understanding. Hence a theory of truth, or truth-
conditions, must occupy a central explanatory role in Davidson's theory
of interpretation.13 A theory of truth not only has the right structure for a
theory of meaning; truth-conditions cash out the kind of understanding
that is appropriate to the notion of meaning, or interpretation, that is
involved in communication.
Given this conception of a theory of linguistic communication, David-
son next argues that the one notion a theory of interpretation should not
employ is the idea of a language common to the interpreter and speaker
if ``a language is anything like what many philosophers and linguists have
supposed,'' namely, a system of ``learned conventions or regularities''
available in advance and independently of the occasion of utterance. In its
place, Davidson proposes a complex account of communication in terms
of speakers' and interpreters' mutual intentions and beliefs. Each S and I
initially brings to each occasion of utterance his own ``prior'' theory of
interpretation. S brings a theory of how he intends his words to be inter-
preted, I brings a theory of how he believes S (most likely) intends his
(S's) words to be interpreted.14 Unsurprisingly, S's and I 's prior theories
From Metaphorical Use to Metaphorical Meaning 41

do not typically match perfectly as they initially stand. S and I therefore

adapt, or mutually adjust, their respective prior theories in order to
achieve the fully shared understanding, or interpretation, of the utterance
that constitutes communication. The resulting theories that evolve in the
course of successful communication Davidson calls ``passing'' theories.
But passing theories, Davidson emphasizes, are too utterance-, occasion-,
interpreter-, and speaker-specic to count as languages. So, if previous
philosophers have posited language only in order to explain how com-
munication by speech is possible, it follows, Davidson concludes, that
``there is no such thing as a language, not if a language is anything like
what many philosophers and linguists have supposed'' (NDE 446).15
This, I would add, is the full force of Davidson's conception of a theory of
interpretation as a theory of use or utterances. Not only the explanandum
is use or utterances; the explanans also need not appeal to anything that
cannot be extracted or abstracted from use.
With this background, let's return to Davidson's notion of literal
meaning, the meanings of words as opposed to what they are used to do.
What work does this notion do in Davidson's language-liberated theory
of interpretation? Indeed how is Davidson able to hold onto the very
notion of literal meaning without falling back on explanatory notions like
language, a notion he considers bogus?
If we think of an utterance as a rational action whose interpretations
are its speaker's communicative and noncommunicative intentions,
Davidson proposes to identify the literal meanings of the words in the
utterance by ordering its interpretations, or intentional descriptions, in
terms of their means-ends relations to each other.16 For example, my
utterance of `yoreid geshem' (in Hebrew) addressed to my son is a means
to saying that it is raining, the saying of which is a means to asking him to
take the laundry o the line, which is a means to reminding him of his
house chores, which is a means to demonstrating my parental authority.
The rst intention in order in this seriesto say such-and-suchDavid-
son labels the rst meaning of the utterance, which is used in turn to
achieve the second and later intentions in the ordering (all of which I'll
call secondary intentions). This notion of rst meaning, Davidson pro-
poses, can serve as a ``preliminary stab at characterizing'' (NDE 434) the
ordinary notion of literal meaning, ``what words mean'' as opposed to
``what they are used to do.''17
Note that Davidson singles out rst meaning as literal meaning only
because it is rst in the speaker's order of means to achieve his ultimate
end. Davidson cannot privilege it as linguistic meaning, because he denies
42 Chapter 2

an explanatory role to language in communication. Nor can he identify

rst meaning with literal meaningas the etymology of `literal' might
suggeston the grounds that it is directly encoded in the words as deter-
mined by rules of language (and not merely conveyed by their use).
Davidson denies that there are any such rules of language.
Yet, even on Davidson's deationary conception of the literal, there
would appear to be another problem in identifying it with rst meaning.
The explanatory power of literal meaning is owing to the fact that it
``can be assigned to words and sentences apart from particular contexts
of use . . .'' (WMM 247). But rst meaningthe understanding of an
utterance that consists in knowledge of its truth-conditionsis context-
dependent to an extreme: It is a feature of ``words and sentences as
uttered by a particular speaker on a particular occasion.'' How can we
reconcile these two claims?
Davidsonian exegesis aside, two kinds of context-dependence should be
distinguished. The rst kind is presemantic: It registers the fact that our
assignments of meanings or interpretations to sounds or inscriptions
(including words as purely syntactic entities) depends on various features
of the context. Consider the familiar task, on hearing some concrete
sound pattern, of determining the semantic description it should be
assigned or by which it should be ``typed.'' I hear the sound pattern `'.
Even knowing that the speaker is speaking English, I must decide whether
what I heard was the rst person indexical `I' or the common noun `eye'
or the armative `aye' or the groan `ai'. In making this judgment, we rely
on all sorts of contextual cuesthe appropriateness of the alternative
types within the immediate string and then within the larger discourse,
our beliefs about the speaker and his intentions, and so on. Now, on
Davidson's picture of interpretation, this task is more general and more
radical. On every occasion of utterance, the interpreter is faced with this
task, and not only with the assignment of one among an alternative
number of types within a language but with the assignment of a ``lan-
guage,'' or a theory of rst meaning, as a whole. To be sure, the inter-
preter brings his prior theory of the speaker to the task, so he need not
start from scratch. However, it would be pure chance, or divine provi-
dence, if his prior theory turned out to match the speaker's own theory
perfectly. Each new, previously unknown proper or common name, as
well as deviations from previous word patterns and malapropisms, is a
potential source of mist calling for revision, leading to a dierent passing
theory. And at each turn of discourse, the task begins anew. Therefore,
the interpreter utilizes every available contextual cue or piece of informa-
From Metaphorical Use to Metaphorical Meaning 43

tion in adapting his prior theory to the interpretation of the ongoing

utterance. So, at this presemantic stage, each assignment of a rst meaning
to an utterance in the resulting passing theory is highly context-sensitive
and utterance-specic.
Presemantic context dependence should be distinguished from a second
sort of postsemantic context-dependence. Having assigned an utterance a
rst meaning, that utterance may then be used for an indenite number of
extralinguistic ulterior purposes or intentions: to warn, promise, deceive,
or threaten. Which of these secondary intentions is realized also depends
on the contextthe speaker's and interpreter's mutual beliefs, intentions,
and expectations. Yet, whichever further intention is attributed to the
speaker, and however the utterance is used, its rst meaning remains,
indierently, as the rst of the means to all these ends. Hence the rst
meaning of an utterance is the meaning it has on all its uses or, more
accurately, regardless of how it is so used.18
Given these two ways in which context can bear on the interpretation
of an utterance, it should now be clear that rst meaning is presemanti-
cally context-dependent but postsemantically context-independent. Con-
text assigns a rst meaning to a given utterance but, once assigned, that rst
meaning remains invariant regardless of the further context-dependent
purposes or intentions for which it is used. In a word, the rst meaning of
an utterance is autonomous of its extralinguistic purposes or uses.19 And
for the same reason (though, as we'll see in ch. 8, it can't be the whole
reason), rst meaning is an interesting candidate for literal meaning: that
is, the meaning of words, words regardless of their use. Here, then, the
criterion of literal meaning is postsemantic context-independence.
With this account of literal meaning in mind, let's return to metaphor.
Davidson's objection to metaphorical meaning cannot be that it is pre-
semantically context dependentbecause all rst or literal meaning is
presemantically context dependent. The objection must be that meta-
phorical interpretation is also postsemantically context dependent: What
the utterance communicates qua metaphor is not autonomous of its
extralinguistic secondary intention and, therefore, cannot be meaning.
But what is the ulterior purpose, or secondary intention, of a metaphor?
Davidson charges that other metaphor theorists start from the ``trite and
true observation'' that ``a metaphor makes us attend to some likeness''
(WMM 247) and then falsely read the likeness into the utterance as its
meaning. For Davidson this eectour being made to notice a resem-
blance or some other salient featureis rather the ulterior purpose dis-
tinctive of metaphor. And like other uses of language to obtain special
44 Chapter 2

eects``assertion, hinting, lying, promising or criticizing'' (WMM 259),

uses that do not introduce special meanings in addition to or in place of
their ordinary literal/rst meaningsno special meaning should be asso-
ciated with the metaphorical utterance beyond its literal/rst meaning.
No explanatory work would be done by such a metaphorical meaning
that is not already done by the combination of literal meanings of the
words uttered and their extralinguistic secondary intention as metaphor to
make us notice a likeness.20
Take, for example, Romeo's utterance of `Juliet is the sun'. Romeo
wants to tell us that and why he loves, even worships, Juliet. He does this
by telling us how she nourishes him with her caring attention, how he
cannot live without her. This he does, not by expressing those thoughts
literally (for whatever reason, e.g., because he doesn't have or know the
right words or because he thinks it will be more eective another way) but
by making us notice them, or by inviting us to discover them, by likening
Juliet to the sun. And this he does, not by literally comparing Juliet to the
sun or by visually displaying Juliet and the actual sun, but by uttering the
(false) sentence `Juliet is the sun'. Of course, for this utterance to make us
consider features that Juliet and the sun share, `Juliet' must mean Juliet
and `the sun' must mean the sunthat is, the words in the utterance must
have their literal/rst meaning. Q.E.D., Davidson argues, we have fully
explained Romeo's utterance by taking into account only the literal/rst
meanings of the words uttered and the special metaphorical ulterior pur-
pose or intention of the speaker in performing the utterance, namely, to
bring features of likeness to our attention. With this explanation in hand,
it would be gratuitous to posit a special metaphorical meaning.
As we saw earlier with Searle's features, this is a cogent argument
against taking the likeness we are made to notice as the meaning of the
metaphor. But the argument leaves untouched the possibility that a meta-
phorical meaning might be something else. Let me restate the argument,
in words Davidson himself would never have put it: Suppose metaphors
are truth-valued, or express propositions or content, and, in particular,
that Romeo's utterance u of `Juliet is the sun' is true i Juliet is P, where
P is a feature u makes us notice in virtue of which Juliet resembles the sun.
Davidson's argument shows that P cannot be the meaning of ueven
though it is a constituent of its truth-condition or propositional content.
More generally, the argument shows that, if there is a metaphorical mean-
ing, it cannot be its truth-condition or propositional content. However,
the argument does not exclude the very idea of a metaphorical meaning.
From Metaphorical Use to Metaphorical Meaning 45

If there is some other candidate for metaphorical meaning, it must sat-

isfy the two conditions I mentioned (pace Davidson) at the end of section
I. (i) It must be a feature we can specify independently both of the par-
ticular utterance in which it occurs and of its context, and (ii) it must be a
feature that explains some regular, systematic feature of metaphorical
utterances. Re (ii): In chapter 1, section II, I presented a number of
examples [(1)(10)] of apparent systematic variations in metaphorical
interpretation; later in this chapter I shall oer additional data. Re (i):
Recall that demonstratives are also expressions whose meanings and
truth-conditions (in a context) must be distinguished and whose mean-
ings, although sensitive to their contexts, can be specied independently
of any particular utterance in a particular context.21 The trick, then, is
to show that there is a rule similar to those that govern the context-
dependence of demonstratives that explains regular and systematic varia-
tions in metaphorical interpretations. Such a rule would constitute a kind
of metaphorical meaning that belongs to our knowledge of semantics
rather than pragmatics.
To get a better grasp of this alternative notion of metaphorical mean-
ing, I want to turn now to the least explicit part of Davidson's story: his
explanation of how a metaphor works simply in terms of its rst/literal
meaning and its metaphorical secondary intention. In particular, I'll focus
on his claim that the utterance of, say, the sentence `Juliet is the sun',
exclusively with its literal meaning, ``makes us notice'' a feature whereby
Juliet resembles the sun in virtue of which (or of the recognition of which)
the metaphor ``works.'' In the standard vocabulary of the metaphor lit-
erature, this part of Davidson's account corresponds to the point at which
``the metaphorical depends on the literal.'' For Davidson this formula
should be understood to say that the metaphorical use of a sentence
depends (exclusively, to the extent to which it depends on meaning) on its
literal meaning. In his writings, Davidson suggests two models for this
dependence. In WMM the model is causal; in NDE it is inspired by
Donnellan's account of the referential denite description. In the next two
sections I'll discuss the inadequacies of these two models to motivate why
we need metaphorical meaning and explain what it must be.

III Metaphorical/literal Dependence I: Davidson's Causal Explanation

Davidson introduces his ``causal account'' of metaphors by comparing

them to jokes, dreams, pictures, and bumps on the head.22 Jokes, for ex-
46 Chapter 2

ample, make us laugh; metaphors make us see likenesses. But no one

would posit a ``joker meaning'' in jokes (in addition to or in place of their
literal meaning) to explain why they make us laugh. For the same reason,
we shouldn't posit a metaphorical meaning in metaphors to explain their
eect. Instead, ``joke or dream or metaphor can, like a picture or a bump
on the head, make us appreciate some factbut not by standing for, or
expressing, the fact'' (WMM 262). Just as the apple that dropped on his
head reputedly caused Newton to think of the First Law of Gravity, so
Romeo's (literally interpreted) utterance of `Juliet is the sun'a spatio-
temporal eventcauses him and us to see (or recognize or notice or
think) that Juliet is like the sun in, say, that we cannot live without either
of them.23 And just as the apple itself neither meant nor served as a rea-
son for believing the First Law of Gravity, neither does Romeo's utter-
ance express that resemblance as its meaning nor furnish a reason for the
belief that Juliet resembles the sun in some respect.
There is also a second parallel between metaphors, pictures, and jokes.
The eect of a joke is a function of its telling in its entirety. A joke cannot
be reduced to, or identied with, any of its isolable parts; just try to pack
a whole joke into its one punch line. Similarly, pictures have no internal
syntax; what the picture as a whole represents is not a function of what its
parts individually represent independently of the context of the picture as
a whole. In whatever way pictures do represent or mean, it is as basic
wholes, not (like language) as composites of separable stand-alone values
of their parts. To some extent the same is true of metaphor, at least on
Davidson's account. Because he acknowledges no metaphorical meaning
that is added to or replaces the literal meaning of one rather than another
constituent expression in a metaphorical utterance, he cannot draw any
distinction between the constituents that would make one or another
subsentential expression ``the'' metaphor. The metaphor is whatever
causes the appropriate eect, whatever makes us notice the ``metaphori-
cal'' feature or resemblance, and this must be the utterance in its totality.
Neither `Juliet' nor `(is) the sun' individually and in isolation would cause
us to see a resemblance between the two of them; hence, the minimal lin-
guistic unit constituting ``the'' metaphorwhat does the alerting, inspir-
ing, promptingmust always be the full utterance in its context. On this
score, metaphors, pictures, and jokes are all alike.
On the other hand, Davidson also claimsindeed this is his primary
positive thesisthat what metaphors achieve ``depends entirely on the
ordinary meanings of those words and hence on the ordinary meanings of
the sentences they comprise'' (WMM 247, my emphasis). ``As much as
From Metaphorical Use to Metaphorical Meaning 47

metaphor as can be explained in terms of meaning may, and indeed must,

be explained by appeal to the literal meanings of the words'' (WMM
256257). But if ordinary literal meanings of words are combined into
complex literal meanings using principles of semantic compositionality,
and ordinary literal meanings of sentences are a compositional function of
the meanings of their parts, then if the metaphorical signicance or eect
of an utterance ``depends entirely'' on its literal meaning, we would think
that it should also be explicable compositionally in terms of the words
and sentences so used. To the extent to which this holds, the analogy with
pictures and jokes breaks down.
To complicate matters even more, note that Davidson shifts from
talking of the ordinary literal meaning of the words that occur in a meta-
phor to that of the sentences that comprise those words. It is clear, as I'll
show, that a metaphorical use depends on the literal meaning of the
word(s) so used, but it is not nearly as clear that it depends on the literal
meaning of the sentence comprised by those words. That depends, I'll
argue, on what in particular we take literal meaning, or literal under-
standing, to be. To begin with, there is the argument that has recently
been put forth (independently) by Avishai Margalit and Naomi Gold-
blum (1994) and by Roger White (1996) who have urged that each of the
individual words in the metaphor `Juliet is the sun' may have, and retain,
its literal meaning, but that the whole sentence does not. For if we iden-
tify the literal meaning of a sentence with its truth-conditions, as does
Davidson, they argue, we do not even understand what must be the case
for that sentence to be true (not just conrmed or known or veried to be
true), namely, the conditions in which Juliet, a human being, is the sun, a
star. Of course, we know that and why the sentence is false, but knowl-
edge of truth-conditions requires that we also know what the world would
have to be like for the sentence to obtainwhich, in this case, we do not.
But if the sentence used metaphorically does not have its ordinary literal
meaning, its metaphorical eect or signicance obviously cannot depend
on or be explained in terms of that meaning.
Davidson might reply at this point that the objection assumes a dis-
tinction between the literally meaningless and the false which, in turn,
presupposes something like the analytic-synthetic distinction.24 To avoid
that controversy, let's therefore allow that `Juliet is the sun' is literally
meaningful and merely false. Yet, in point of fact, Davidson's own ex-
planation of how metaphor works does not appeal to more than the sep-
arate literal meanings of the individual words in the sentence, ignoring
any contribution made by the string syntactically or semantically struc-
48 Chapter 2

tured as a sentence. For Davidson there is no dierence between a meta-

phor and a poem like T. S. Eliot's ``The Hippopotamus'': both are
``devices that alert us to aspects of the world by inviting us to make
comparisons.'' But Eliot's poem works simply by the alternating pre-
sentation or displaythe brute juxtaposition, as it wereof stanzas or
clauses referring to hippopotami and the Church. Likewise, Davidson
would have us believe that metaphor works simply by way of the linear
sequence of literal meanings of the individual words of the utterance, re-
gardless of its sentential syntax. Although the metaphorical depends on
the literal, there is nothing specic to the literal meaning of the sentence,
beyond that of its component words, that does any work in Davidson's
account of the dependence of a metaphor on the literal.
Our knowledge of the separate and separable literal meanings of the
individual words that make up a string cannot, however, be sucient to
explain how that string works as a metaphor. If the co-occurrence (as in
Eliot's poem) of the words in the utterance with their literal meanings is
enough to account for the metaphorical eect, then sentences that dier
only, say, in their respective subject-predicate structuresfor example, `a
man is a wolf ' and `a wolf is a man'should not have systematically
dierent eects. If we can ignore their (literal) sentential syntactic dier-
ences because they both invite us to make comparisons between the same
things or words, the two sentences should not dier predictably in their
metaphorical uses.25 It should also be emphasized that the dierence be-
tween `a man is a wolf ' and `a wolf is a man' is not only a dierence in
linear surface word order, but a dierence between their respective syn-
tactic structures, and this dierence lies at a deeper, more abstract level of
representation of the sentence. If we include an account of these repre-
sentational dierences in the explanation of the metaphorical eect, it
follows that the eect will depend not just on the literal meanings of the
constituent words but also on some aspect of the literal meaning of the
sentence used, in particular, the relevant kind of meaning, or logical form,
determined by or corresponding to its syntax. This will not be captured
simply by truth-conditions; it requires an additional notion of sentential
meaning.
Let's take stock. I have argued that if we understand the formula ``the
metaphorical depends on the literal'' to mean, following Davidson, that
the metaphorical use of a sentence depends (exclusively) on its literal
meaning, then: (i) the literal meanings of the individual words so used do
carry part of the explanatory burden; (ii) (pace Margalit/Goldblum and
White) the literal truth-conditions of the sentence uttered play no explan-
From Metaphorical Use to Metaphorical Meaning 49

atory role; but (iii) certain syntactic or logical-formal properties of the

sentence do contribute to or constrain the metaphorical use. For reasons
that will emerge in the next two sections, I also think that (ii) holds, not
only for the literally false or anomalous metaphors Margalit/Goldblum
and White discuss, but also for metaphors that are literally true and ac-
ceptable. Now, the shape of the kind of literal meaning structure that
meets (i)(iii) is not easy to discern (although we shall see in ch. 4 that our
metaphorical characters t it). One tentative conclusion we can draw is
that literal (sentential) meaning is not identical with truth-conditions.
However, the more important lesson for our immediate purposes is that
the combination of (i)(iii) is, in any case, not what we would expect on
Davidson's causal model based on his conception of interpretation as on-
the-spot construction of passing theories of utterances in their contexts, as
opposed to xed theories of language or grammar that are endowed or
learned prior to and independently of the utterance. If causes are singular
events, one would expect the kind of literal meaning on which the meta-
phorical eect depends to be something to be read o, as closely as pos-
sible, of the actual concrete utterance. The two natural candidates would
be either the literal truth-conditions of the utterance or the literal mean-
ings of its individual words. The least natural candidate for Davidson's
needs would be the more abstract representation delineated by (i)(iii), a
structure that requires knowledge not just of the literal meanings of the
constituent words but also of deeper logical form abstracted away from
the truth-conditions of the utterance. So, if (as Davidson holds) meta-
phorical use depends on literal meaning, the kind of literal meaning that
must be taken into account does not square with Davidson's picture of
occasion-specic interpretation and use.
Before going on, and in part to prepare the way for my own account, it
is worth pausing to consider the ``literal meanings'' (of words) to which
Davidson repeatedly refers. What are they? More specically, what must
literal meanings be in order to do the explanatory work Davidson wants
them to do for metaphor?26 I have no positive answer to the question, but
two candidates can be excluded.
Recall that Davidson makes the metaphorical use of a sentence ``de-
pend entirely on the ordinary [literal] meanings of those words and hence
on the ordinary [literal] meanings of the sentences they comprise'' (my
emphasis). That is, literal meaning is compositional: if two expressions F
and C have the same literal meaning, then . . . C=F . . . (the sentence that
results from substituting C for all occurrences of F in . . . F . . .) has the
same literal meaning as . . . F. . . . Now,
50 Chapter 2

(1) Juliet is the sun

interpreted literally is an extensional sentence that is true if and only if
Juliet is the sun. Furthermore, any sentence (10 ) that would result by
replacing, say, `the sun' by another expression with the same literal mean-
ing ought to have the same literal meaning as (1). Therefore, if the meta-
phorical use or eect of an utterance of (1) depends entirely on its literal
meaning, and (10 ) has the same literal meaning as (1), its metaphorical use
ought to be preserved.
What could be such a literal meaning of a word on which its meta-
phorical eect, or interpretation, depends? What kind of literal meaning
would preserve the metaphorical interpretation of its literal vehicle under
substitution of a literally synonymous vehicle? A rst candidate would be
its extension or referent. However, consider Romeo's utterance of (1) in
the context c in which `is the sun' is used metaphorically to express the
property of being worthy of worship (or, equivalently, makes us notice
that Juliet, like the sun, is worthy of worship). Suppose also that (1), so
used, is true in the circumstance of that context c: Juliet is worthy of
worship. And suppose that
(2) The sun the largest gaseous blob in the solar system
interpreted literally, is also true. Hence the two anking terms (literally)
co-refer and, by hypothesis, have the same literal meaning. Yet, from (1)
and (2) it does not follow that
(3) Juliet is the largest gaseous blob in the solar system
is true even where `the largest gaseous blob in the solar system' is also used
metaphorically in the same context c; for the two literally coextensive
terms, each used metaphorically in the same context, may none the less
express metaphorical interpretations or make us notice metaphorical
eects that are not themselves coextensive in any worlds in which they are
simultaneously evaluated. Hence, the two (metaphorically used) sentences
will not necessarily both be true even when they are both evaluated in the
world in which they are both uttered. It follows that the metaphorical use/
interpretation of an utterance cannot depend entirely, or solely, on the
literal meanings of its words so long as a literal meaning is an extension
and we individuate uses, eects, or interpretations (minimally) by mate-
rial equivalence.
A second natural candidate for the literal meaning of a word would be
(despite Davidson's extensionalism) its intension (or some extensional-
istically acceptable surrogate).27 But by the same reasoning, if metaphor
From Metaphorical Use to Metaphorical Meaning 51

depends on literal meaning, any two expressions with the same intension,
or one expression with a constant intension, should have the same meta-
phorical eect or interpretation on all occasions and in every context.
This is not generally true. Metaphors, as we saw in examples (1)(10) of
chapter 1, vary systematically with their contexts. In some cases, the in-
terpretation of the metaphor seems to vary with linguistic features of its
context, and in other cases, with nonlinguistic features. For example,
`sweet' metaphorically applied to words (as in `sweet words') expresses the
property of being pleasant for speakers of English, Hebrew, and many
other languages, but for speakers of Chinese the corresponding transla-
tion expresses the property of being specious. Calling one's lover `a bed-
bug' in American English is calling him or her a nuisance or pain, but in
Nigeria the term is one of aectionapparently because bedbugs there
(as opposed to everywhere else) are thought to be cute.28 This dierence
in metaphorical interpretation or eect is due to cultural dierences be-
tween the respective speech communitiesdierences in (socially shared)
presuppositions and beliefs. But there seems to be no way of incorporat-
ing such factors into an account framed only in terms of intensionality.
Intensions (or some feature ner than extensions) may indeed turn out
to be necessary but a purely intensional account of metaphor neglects
another factor essential to its interpretation: the role of its context. So, if
we take ``literal meaning'' to mean either an intension or some exten-
sionalistically acceptable surrogate, Davidson's claim that metaphorical
use ``depends entirely'' on literal meaning should still not be taken liter-
ally: Somehow the account must incorporate the role of context.29
My last point concerns Davidson's claim that there is a causal expla-
nation of how the metaphorical eect of an utterance, the feature we
subsequently notice or, more accurately, the noticing of the feature,
``depends on its literal meaning.'' Davidson insists (WMM 249) and I
agree (cf. ch. 1, sec. III (ii)) that any account that fails to explain this
metaphorical-literal dependence is inadequate as an explanation of meta-
phor. But it is also Davidson's view of causation and explanation (and I'll
grant this for the sake of argument) that a singular causal statement like
(4) Romeo's utterance of `Juliet is the sun' caused his and his audience's
noticing that Juliet, like the sun, is worthy of worship
is explanatory only because, and insofar as, it implies the existence of a
law that covers the case. I shall argue now that, from the little Davidson
says about the causal explanation of the metaphorical eect, there is no
reason to think that such a law would (and perhaps good reason to think
52 Chapter 2

that it would not) include the required condition that the metaphorical
eect or use depends on the literal meaning of the words or sentence. In
that case, Davidson does not oer an explanation of metaphor that would
be sucient on his own grounds. Hence his own use account does not
thereby render unnecessary the rival account oered by a semantic theory
of metaphor that does explain metaphorical-literal dependence. At most
Davidson has given us singular causal descriptions of our metaphorical
utterances and their eects. But those descriptions of use are compatible
with a semantic explanation of metaphor.
Before spelling out the problem for Davidson, I want to show how the
same diculty arises on a more radical version of his position presented
by Richard Rorty (1987). Rorty enthusiastically endorses Davidson's
oering of a fully naturalistic but noncognitivist alternative to the re-
ceived view of ``the tradition'' that explains why metaphors are ``indispen-
sable'' by forcing them into a cognitivist mold in which they express a sui
generis kind of meaning. Davidson ``lets us see metaphors on the model
of unfamiliar events in the natural worldcauses of changing beliefs
and desiresrather than on the model of representations of unfamiliar
worlds.''30 However, in order to expose metaphors as nothing more than
noncognitivist causes``mere stimuli, mere evocations'' (ibid., 291)
Rorty goes considerably further than Davidson. He divests them of their
status, not only as bearers of special metaphorical meaning, but also as
uses of language and even as intentional, rational actions.
Adopting a gure from Quine (1978), Rorty characterizes the ``realm of
meaning,'' the domain of semantics, as ``a relatively small `cleared' area
within the jungle of use.'' In this clearing there ourishes the ``regular,
predictable, linguistic behavior'' that constitutes the literal use of lan-
guage. Metaphorwhich, according to Rorty, is unpredictable and
irregular by its very naturefalls outside the scope of meaning, in the
surrounding ``jungle of use'' (Rorty 1987, 285), a jungle populated by
unfamiliar birdcalls, thunderclaps, and the whole spectrum of exotic
sounds found in nature as well as by (some) speech utterances. What all
these sounds have in commonand what apparently makes them a jun-
gleis that they are all just noise. They are sounds for which we have no
(Davidsonian) prior theory to make sense of them; instead, by their very
strangeness and anomalousness, they simply cause eects that impinge on
us. Metaphors are also such more or less exotic noise. They may stand a
tat closer than birdcalls to the cleared space of literal meaning because
diachronically they can dieor, as Rorty puts it, be ``killed o ''and
live on in the afterlife of ``stale, familiar, unparadoxical, and platitudi-
From Metaphorical Use to Metaphorical Meaning 53

nous'' literal language (ibid., 295). But as long as they are genuinely alive,
metaphors are nothing more (nor less) than noise. Given their irregularity
and unpredictability, they cannot be ``understood'' or ``interpreted''
except in ``the way that we come to understand anomalous natural
phenomena'' (ibid., 290), namely, by revising our antecedent (scientic)
theories. So, while metaphors may be ``responsible for a lot of cognition,''
Rorty says that this is only in the nonintentional causal sense of ``re-
sponsible'' in which ``the same can be said about anomalous nonlinguistic
phenomena like platypuses and pulsars'' (ibid.). Our ability to ``under-
stand'' a metaphor does not, in short, fall under the rubric of mastery of
language. Indeed the causal power of a metaphor also does not depend on
any of its properties as language or even as a species of intentional action.
Like a thunderclap or birdsong, the utterance of a metaphor is nothing
but a nonintentional event, despite all its causal functioning.
Davidson might also locate metaphor in a ``jungle of use,'' but the
jungle would not be Rorty's. Davidson begins, as I noted earlier, with
actual utterances or linguistic acts. The pretheoretical state of their
``meaning'' might be described as a ``jungle'' because it is dense, undif-
ferentiated, and only after considerable theory-laden pruning and trim-
ming, systematically dierentiable into the meanings of its words (their
literal or rst meaning) and the various secondary meanings that consti-
tute all the other things words can be used to do. However, even the un-
ruly jungle that remains after we have cleared away the space of literal
meaning is populated only by the meanings of intentional, rational lin-
guistic actions. Some of the secondary extralinguistic ends may be causal
eects of the utterancesas with metaphorsbut the utterances are no
less rational and intentional because they also have causal eects. For
Davidson, unlike Rorty, it is never noisesnonlinguistic, nonintentional
eventsthat fall in our domain of investigation, only intentional human
actions, of which linguistic actions are one subclass. Within that corpus
we distinguish what words mean, the subject of semantics, and the many
other things they are used to do or mean, including (as in metaphor) those
things that their use causes us to recognize or see.
Not all jungles, then, are created equally wild. But despite their dier-
ences, Rorty's picture drives home more perspicuously than Davidson's
one important consequence of a ``causal'' explanation of metaphor. The
fact that the utterance of a metaphor does (some) causal work is not in-
compatible with its being an intentional linguistic action, but there is little
if any reason to expect that a causal explanation of metaphor, using
(strict) laws, will make reference to its character as an intentional act or as
54 Chapter 2

a use of language. To the extent to which a metaphor's causal explanation

and underlying laws all treat it as a physical event, Rorty is correct to
view metaphors as just noise, that is, nonintentional, nonlinguistic events.
But this consequence amounts to a reductio argument against Davidson's
positionbecause metaphors are not just noise. As everyone (including
Davidson) agrees, a metaphorical interpretation, or eect, depends on its
literal meaning. But noise, being nonlinguistic, has no literal meaning. So,
Rorty is wrong to describe metaphors solely as noise because it is not
possible to explain them fully as noise. What his story enables us to see
sharply is that a Davidsonian causal account cannot claim both to give
a causal explanation of metaphor and to explain how the metaphorical
eect ``depends entirely on the ordinary [i.e., literal] meanings of those
words.'' For the causal explanation of a metaphor will treat it as a non-
intentional, nonlinguistic eventas meaningless noisewithout men-
tioning features like its literal meaning. Or, to be a bit more cautious,
there is no reason to think that a causal explanation of the metaphorical
eect employing (strict) laws will employ concepts like that of literal
meaning.
Let me recast this argument from a dierent angle. Recall that on
Davidson's own account of causal explanation, for one event to be said to
cause another it is only necessary that the events be nomologically related
under some description of each.31 Furthermore, a singular causal state-
ment need not itself instantiate the causal explanatory law that it pre-
supposes or refer to the events in question under their nomologically
appropriate description.32 Hence we can agree with Davidson that the
metaphor-utterance event causes the resemblance-noticing event (even
allowing for the anomalism of the mental) insofar as there are grounds to
believe that there exists some nomological relation between the two under
some description of each. But this is not enough for Davidson's account
of metaphor. His claim is not simply that there is some causal relation
between the metaphorical utterance and what it makes us noticeunder
some appropriate description of each. Davidson's thesis is that we can
explain how the causal ``eect'' of the metaphor ``is brought o by the
imaginative employment of words and sentences [and depends ] entirely on
the ordinary meanings of those words and hence on the ordinary mean-
ings of the sentences they comprise'' (WMM 247, my italics). The vo-
cabulary of the explanation must itself, then, make reference to words,
use, and literal (or rst) meaning. But it remains to be shown that any
such explanation can plausibly be given. Davidson, for one, gives no evi-
dence that such an explanation is in the ong.
From Metaphorical Use to Metaphorical Meaning 55

Let me emphasize that I am not objecting to the ultimate possibility of

causal explanations of metaphor or, for that matter, of any linguistic
phenomenon in terms of the physical or neurophysiological properties of
utterances and their eects. Such an explanation would, in principle, treat
the utterance of a metaphor and its eect no dierently than the auditory
perception of a thunderclap or a birdsong and their eects on us. I am
objecting instead to the claim that we can explain how a metaphor works
both as a use of language that depends entirely on the literal meanings of
its words and in terms of a causal relation to the eect of its utterance. We
may be able to make the singular judgment that Romeo's utterance
caused us to see that we can't live without Juliet ( just as we can't live
without the sun), but as soon as we try to provide an explicit causal ex-
planation of that eect, we lose the substance of Davidson's claim that it
is as a use of language exclusively with its literal meaning that metaphor
does its work. On the other hand, a singular causal judgment is not itself
incompatible with a semantic explanation employing a notion of meta-
phorical meaning. To rule out the latter, Davidson would have to provide
a causal explanation that renders the notion of metaphorical meaning
unnecessary. However, such an account would require a lawlike explana-
tion of metaphor as a kind of use of the sentence with its literal meaning
in which the explanans itself employs literal meanings, words, and so on.
But then metaphors would not only be noises or bumps on the head
unless those also bear meanings in the head.

IV Metaphorical/literal Dependence II: Davidson on Referential Denite

Descriptions, Malapropisms, and Metaphor

While contrasting malapropisms with Keith Donnellan's (1966) well-

known idea of a referential denite description, Davidson writes:
Jones has said something true by using a sentence that is false. This is done in-
tentionally all the time, for example in irony or metaphor. A coherent theory
could not allow that under the circumstances Jones' sentence was true; nor would
Jones think so if he knew the facts. Jones' belief about who murdered Smith can-
not change the truth of the sentence he uses (and for the same reason cannot
change the reference of the words in the sentence). (NDE 440)
Here Davidson claims that when a speaker uses a denite description
referentiallyas opposed to attributivelyhe succeeds in communicating
his intended meaning (or reference) despite the fact that it diers from the
literal/rst meaning (or reference) of the words. Yet the words have their
literal/rst meaning on that use. Hence the referential-attributive distinc-
56 Chapter 2

tion marks a distinction between two uses of language that dier in ``what
they say''that is, in their truth-conditions and, therefore, in truth-
valuewithout diering in their literal/rst meanings. Davidson con-
trasts this case with malapropisms (and the introduction of new proper
and common names) where the literal/rst meaning of the word used itself
changes in its context of utterance. Metaphor, Davidson adds almost as
an aside, is like the use of the referential denite description. The sentence
is used metaphorically to say something true, although the sentence used
is nonetheless (literally) false; hence the sentence retains its literal/rst
meaning. Thus metaphor, like the referential description, diers from
malapropism, whose literal/rst meaning undergoes a change on that use
from its prior literal/rst meaning.33
To support his account of the referential use of a denite descrip-
tion, Davidson recounts Donnellan's well-known example of Jones who
believes, though (unbeknownst to himself ) falsely, that Max murdered
Smith. Seeing Max's behavior in the courtroom dock, Jones exclaims:
`Smith's murderer is insane'. Where Jones uses the description `Smith's
murderer' referentiallyas a tool to refer to Max but not to describe him
Donnellan argues, and Davidson agrees, that Jones says ``something
true'' provided Max is insane.
In this example there is no reason to say that the referential denite
description has undergone a change of meaning just because Jones is able
to ``say something true'' about Max by uttering `Smith's murderer is
insane' even though Max does not satisfy the content of the description.
On the contrary: Jones is able to use the description `Smith's murderer' to
refer successfully to Max because he (falsely, as it turns out) believes, or
represents himself as believing, that Max is Smith's murderer in the prior,
literal meaning of those words. And it is either because his interpreter also
(falsely) shares that belief or because he recognizes that Jones (falsely)
holds that belief of Max (in the prior, literal meaning of the words) that
he interprets Jones' utterance of `Smith's murderer is insane' to be saying
something true about Max. Were we to inform Jones that Max is not in
fact Smith's murderer (in the prior, literal meaning of the words), he
would not respond that we had failed to understand the meaning of his
use of that description on that occasion, or that he had intended the
words to have a new passing rst meaning in that context, a meaning
according to which they did uniquely designate Max. Instead he would
saygiven his intention to use the description referentiallythat he had
meant to say something true about Max even though he was wrong in
believing that he was Smith's murderer (in its passing literal meaning),
From Metaphorical Use to Metaphorical Meaning 57

hence wrong (or misleading) in using that description. So, as Davidson

concludes, ``the reference is nonetheless achieved by way of the normal
meanings of the words. The words therefore must have their usual refer-
ence'' (NDE 439).
With a malapropism, in contrast, Davidson claims that the word itself
acquires as its (passing rst) meaning what the speaker intends to say in
its context of utterance. For example, when Yogi Berra reputedly thanked
the crowd in Yankee Stadium on ``Yogi Berra Day'' for ``making this day
necessary,'' possible is what the word `necessary' meant on that occasion,
and not merely what Yogi meant by using that word (in its ordinary, lit-
eral meaning).34 Here is how Davidson would describe this situation.
Yogi's interpreter brings a prior theory of Yogi's linguistic behavior to the
occasion of utterance, a theory in which `necessary' means necessary.
Hearing the absurdity or inappropriateness of Yogi's utterance with that
prior rst meaning, he knows that Yogi must have intended to say some-
thing else. At this point the interpreter has at least two options. Either he
can keep its prior rst meaning as its passing rst meaning and add an
ulterior purpose for the utterance, say, take Yogi to be speaking ironically
or comically; or he can adjust his prior theory of Yogi's rst meaning in
order to give the word `necessary' as its passing rst meaning what he
believes Yogi intends to be saying on that occasion, namely, possible.
Davidson does not tell us how Yogi's interpreter decides to interpret the
utterance one way rather than another. However, if he decides that Yogi
intends to be saying that he wants to thank everyone for making this day
possible, then it is arguable that there is at least as good a (if not a better)
reason to take possible as the passing rst meaning of `necessary' as there
is to take it as something imparted secondarily by Yogi's use of `neces-
sary' in its prior rst meaning. Unlike the case of a referential description,
where there are grounds for saying that Jones understands his words in
their prior rst meaning (namely, the fact that he believes, albeit falsely,
that Max is Smith's murderer, in the prior rst meaning of these words),
here there is no such reason to think that Yogi understands by `necessary'
necessary, that he intends to say that he wants to thank everyone for
making the day necessary in its prior rst meaning. On the contrary, we
would explain why he said what he did by saying that he intended to
thank everyone for making the day possibleand he believed, or repre-
sented himself as believing, that on this occasion `necessary' just (rst)
means possible. To be sure, Yogi's interpreter can gure this out only be-
cause he knows that in Yogi's prior theory `necessary' means necessary
and `possible' means possible, and he conjectures that Yogi is ``confusing''
58 Chapter 2

the two (or that they get confused in his mental lexicon because of their
semantic proximity). However, that the interpreter knows one interpre-
tation ``by way of '' the other does not show that the latter belongs to
his passing theory and should be prior to Yogi's intended interpreta-
tion of the word in the passing theory itself. For the passing theory is a
theory of what the speaker intends on the occasion of utterance. And
Yogi had no such intentionto mean necessary by `necessary'on that
occasion.
For the sake of argument, let's grant Davidson these characterizations
of referential denite descriptions as uses of words to mean something
other than what the words themselves literally/rst mean and of mala-
propisms as words that acquire a new passing literal/rst meaning on
their occasion of utterance. How should we then describe metaphor?
Davidson gives no explicit argument for his classication of metaphors
with referential denite descriptions in the realm of use, rather than with
malapropisms in the realm of meaning. But his reasoning would presum-
ably go like this: What a metaphor makes us notice ``depends on its
ordinary [ literal] meanings'' (WMM 247); therefore, ``an adequate
account of metaphor must allow that the primary or original [ literal]
meanings of words remain active in their metaphorical setting'' (WMM
249). Hence a metaphor like `Juliet is the sun' must ``have'' its ordinary,
primary, or original meaningthat is, its prior literal/rst meaningas
its passing literal/rst meaning on its occasion of utterance. In this re-
spect, metaphor is like a referential denite description and unlike a mal-
apropism. The speaker uses a literal/rst meaning in order to make us
notice hitherto unnoticed resemblances or features, his ulterior purpose
for his metaphorical utterance.
But is the way in which a metaphor ``depends'' on its literal/rst mean-
ing really like the way in which a referential denite description ``depends''
on its prior literal/rst meaning? How do we know that ``the primary or
original meanings of words remain active in their metaphorical setting''
within the passing theory of the utterance? Perhaps the metaphorical-
literal dependence holds between a rst meaning that the metaphorical
utterance acquires in its passing theory and the previous rst meaning
the word held in its prior theory?
The answer to these questions is, I think, that we have no satisfactory
answer using just the conceptual resources of Davidson's theory. We lack
for metaphorical utterances the motivation we had for referential denite
descriptions to maintain that their prior literal/rst meaning remains their
passing literal/rst meaning. Recall that Jones must intend his words
From Metaphorical Use to Metaphorical Meaning 59

`Smith's murderer' to be understood according to their prior literal/rst

meaning because he believes that Max is Smith's murderer in their prior
literal/rst meaning, a belief we must ascribe to him to explain why he
uses the (improper) description `Smith's murderer' to refer to Max. There
is no analogous reason to claim that the speaker of a metaphor intends his
words to be understood, and hence interpreted, according to their prior
literal/rst meaning. To reinforce this point, recall our distinction between
the literal meanings of the individual words in the metaphorical utterance
and the literal meaning, or truth-conditions, of the sentence so used.
Focus rst on the sentence. Romeo surely does not intend to communi-
cate the absurd or patently false proposition that Juliet is the sun
expressed by the literal meaning of the sentencewhen he utters `Juliet is
the sun'. As we said, following the argument of Margalit and Goldblum
(1994) and White (1996), a literally deviant string like this may not even
have a literal meaning or literally express a proposition, where either of
these is a set of truth-conditions. Even where the sentence used meta-
phorically is not absurd or false taken literallyfor example, Cohen's
twice-true `no man is an island'there is also no evidence that the literal
meaning of the sentence does any work in explaining the metaphorical use
in the passing theory. Therefore, we have no reason to hold that the
speaker believes (or holds some other attitude toward) that proposition,
hence, that he understands the metaphorical utterance by way of taking
the sentence according to its prior literal/rst meaning in the passing
theory. And in that case, there is also no reason to say that the prior
literal/rst meaning of the sentenceits prior literal truth-conditions
should also be considered its passing rst meaning on occasions when it is
used, or interpreted, metaphorically.35
On the other hand, there is also good reason not to consign the literal
meaning of the words used metaphorically entirely to the prior theory of
the utterance. The literal/rst meaning of the words must ``remain active
in the metaphorical setting.'' Whatever turns out to be the exact nature
of the dependence of the metaphorical on the literal, knowledge of the
literal/rst meaning of words is necessary on the occasion of utterance
itself in order to determine what they metaphorically make us notice, that
is, their metaphorical interpretation. And the relevant literal meaning is
not what the literal meaning of the word was but what it ison the
very occasion on which it is interpreted metaphorically. Therefore, we
shouldn't explicate metaphorical/literal dependence diachronically, hold-
ing between passing and prior theories. Within the passing theory there
must also be room for the literal/rst meaning of the words of the
60 Chapter 2

utterance. The problem for Davidson's account is how to make room for
this in the passing theory itself.
Suppose the metaphorical use depends on the literal/rst meanings of
the words used in the passing theory. Since it is the very nature of literal
meaning to be compositional, those literal word-meanings should be
composible into literal sentence-meanings. But the literal meaning of a
sentence is its truth-conditions, the content of our understanding of the
utterance. Therefore, where the literal meanings of the words are the rst
meaning of the passing theory of the metaphorical utterance, we are also
forced to say that the interpreter takes the speaker to understand his
utterance according to that literal meaning. But that conclusion, we have
argued, is either false, if metaphorical utterances do not have literal
meanings qua truth-conditions, or it lacks motivation, insofar as meta-
phors are not explained in terms of their literal sentence-meanings, how
they would be literally understood.
To put it a bit dierently, I am arguing that the words as used, or
interpreted, metaphorically must ``have'' their literal/rst meaning as part
of their passing theory of interpretationbecause knowledge of their lit-
eral meaning is necessary for knowledge of their metaphorical use or in-
terpretation. But the appropriate sense of ``having their literal meaning''
is not one that can be articulated with the conceptual apparatus of
Davidson's theory. For the words do not ``have'' their literal meaning
in the sense that the sentences they comprise are, or are intended to be,
understood according to their literal/rst meaning, that is, their truth-
conditions. What we therefore need is a way to make sense of how indi-
vidual words can ``have'' their literal meaning without thereby claiming
that the sentence composed of those words used, or interpreted, meta-
phorically must also be understood according to its literal meaning or
truth-conditions.

V The Autonomy of Meaning and Metaphor

I suggested in chapter 1 that demonstratives (including indexicals) should

serve as a model for our semantic theory of metaphor as a type of
context-dependent interpretation of language. As a case in point, I now
want to suggest that demonstratives can also point us toward a way out of
the impasse to which the argument has led in the last two sections: a way
to make sense of the idea that a metaphor may ``depend'' on the literal
meaning of its words but not on the literal meaning (i.e., truth-conditions)
of the sentence those words comprise. This way out of the impasse will
From Metaphorical Use to Metaphorical Meaning 61

also direct us away from Davidson's conception of meaning in terms of

interpretation in the form of truth-conditions whose shared understanding
constitutes communication, and toward the notion of meaning as char-
acter, which will feature in Kaplan's semantics in chapter 3.
As we have already had occasion to note, in his less cautious moments
Davidson writes as if ``literal meaning'' and ``literal truth-conditions'' are
equivalent and that what gives both of them ``genuine explanatory
power'' is that they ``can be assigned to words and sentences apart from
particular contexts of use'' (WMM 247). This, of course, is not always
the case, as demonstratives amply show; and Davidson knows this. The
indexical (type) `I' and pure demonstrative (type) `that' have ``literal
meanings'' assigned to all of their utterances (tokens) ``apart from partic-
ular contexts of use,'' but their contributions to the truth-conditions of the
sentences in which they occur vary with, and depend on, their particular
contexts. For example, my, JS's, utterance of `I love Jerusalem' is true i
JS loves Jerusalem, while DD's utterance of the same sentence is true i
DD loves Jerusalem. Yet our two utterances of `I' have the same literal
linguistic meaning, namely, the rule
(I) Each utterance of `I' directly refers to its speaker36
and it is assigned to the utterances ``apart from [their] particular contexts
of use.''
From the beginnings of his program, Davidson has acknowledged the
diculties posed by demonstratives for a Tarski-style theory of truth that
treats sentences simpliciter as the truth-vehicles. To remedy the diculties
he has proposed to make utterances the truth-bearers and to characterize
truth for a language relative to a speaker and time (and possibly other
contextual devices corresponding to other demonstratives).37 Thus in
place of the Tarskian T-sentence (for sentences s containing no demon-
strative expressions)
(T) s is true i p,
which, according to Davidson, captures what a speaker knows when he
knows the meaning of s, he has proposed for the sentence s* (containing
demonstratives, such as `. . . I . . .') a relativized T-sentence of the form
(Td ) s* is true as spoken by a speaker S at a time t i p*
where p* is a translation of s* into the metalanguage and contains the
corresponding contextual value for each occurrence of a demonstrative in
s*. For example:
62 Chapter 2

`. . . I . . . now . . .' is true as spoken by JS on 7/20/98 i JS

7/20/98 .
This treatment of demonstratives raises a number of technical questions
I shall not pursue here.38 What is more critical for our purposes is how
to describe the context-dependence of demonstratives like `I' in a way
compatible with Davidson's criterion that explanatory power requires
context-independence.
Earlier I distinguished presemantic and postsemantic brands of
context-dependence. The context-dependence of `I' is neither. It is not
postsemantic because what varies with context in its case is not an
extralinguistic intention or ulterior purpose performed by the utterance,
but its communicative intention, its truth-condition, the shared under-
standing necessary for communication. The context-dependence of `I' is
also not presemantic. What varies with context in its case is not the lexical
description or type of a (semantically uninterpreted) sound, shape, or
word. The sound `' has already been assigned the lexical type of the rst
person demonstrative `I'; indeed it is only because it has already been
assigned that type that it is context-dependent in this further way. There-
fore, in addition to pre- and postsemantic context-dependence, let's dis-
tinguish a third kind of context-dependence for demonstratives: semantic
context-dependence. Here the context functions not to pair a type with an
uninterpreted sound or shape, but rather, given a meaningful type already
assigned to the sound or shape, to determine a truth-conditional factor on
a particular occasion of utterance.
To accommodate demonstratives within his theory of meaning,
Davidson must allow for this type of semantic context-dependence. His
theory of literal meaning, that is, that meanings are truth-conditions,
must be extended to acknowledge that dierent tokens of one sentence
type with one literal meaning (e.g., a sentence containing demonstratives)
can have dierent truth-conditions in dierent contexts; hence not all
truth-conditions can be ``assigned apart from particular contexts of use.''
Yet, the point of context-independence for meaning is satised no less by
the context-sensitive meaning of demonstratives than it is by the meanings
of eternal expressions. A rule like (I) enables us to capture what is regular
and systematic about dierent utterances of the word `I' and we can
specify (I) apart from any particular utterance in which `I' occurs. This
much Davidson himself ought to readily acknowledge. However, having
taken this (small) step, Davidson also ought to be more sympathetic to a
semantics for metaphor if (and, I admit, this is no small ``if '') it could
also be shown that there is a strong parallel between the systematic way
From Metaphorical Use to Metaphorical Meaning 63

sentences containing demonstratives vary across contexts in their truth-

conditions and the way metaphors likewise depend for what they say, for
their content or truth-conditions, on their contexts. The same semantic
theory should cover metaphors no less than demonstratives. Davidson's
charge that metaphorical meaning is ``not a feature of the word that
the word has prior to and independent of the context of use'' would,
then, carry no more weight against metaphor than against semantically
context-dependent expressions like demonstratives.39
The relation between the meaning of a demonstrative, such as (I) for
`I', and its truth-condition(-al factor) also suggests a way to explain how
the metaphorical use of an expression F ``depends'' on its literal mean-
ingthough this explanation will now force us to rethink the notion of
meaning in non-Davidsonian terms. The rule (I) plays a number of roles.
First, it makes it possible to generate an interpretation for the indexical,
the individual referent that is its truth-conditional constituent. However,
that interpretation is underdetermined by the rule of meaning. In dierent
contexts, depending on who is speaking, the same rule of meaning will
yield dierent referents as interpretations. So, the sense in which the in-
terpretation of `I' ``depends'' on its meaning cannot be that the latter
(fully) determines the former. Second, the rule of meaning (I) constrains
the possible interpretations of `I'. No matter who a speaker believes him-
self to be, no matter whom he intends to refer to by his utterance of `I', no
matter how many cues he gives his interpreter about his referential inten-
tions, his utterance of `I' will refer to himself, the actual speaker, and to
no one else. Thus, what one knows by the meaning of `I' is, in part, an
instance of the autonomy of meaning: the fact that what expressions (lit-
erally) mean is independent, or autonomous, of what individual speakers
intend to communicate on given occasions.40 (I) thereby excludes impos-
sible interpretations of `I' rather than determines what is communicated
or expressed by it. In this sense, the interpretation of `I' does depend on its
meaning; the one is constrained by the other.
Knowledge of this kind of meaning, exemplied by (I), is not un-
related to the knowledge of meaningactual truth-conditionson which
Davidson focuses, but it is dierent in function. For Davidson, meaning is
what is understood, and it is shared understanding that makes for infor-
mative communication. Hence meaning is cashed out by truth-conditions,
for what one must know to understand an utterance is (at least) to know
the conditions under which it would be true. But the kind of meaning
expressed in rules like (I) is not itself a truth-condition. The descriptive
condition in (I) is never itself a component in the truth-conditions for its
64 Chapter 2

utterances. Instead the individual referent is the truth-conditional con-

stituent, though (I) and not the individual referent is what the interpreter
knows as the literal meaning of each utterance of `I'. Rather than be-
longing to what is communicated or understood or intended by the utter-
ance, the (I) kind of meaning constrains the possible truth-conditional
factors for its utterances. But for this reason it also does not t neatly into
Davidson's means-ends ordering of rst and secondary intentions. If rst
meaning is how the utterance is primarily understood, and intended to be
understood, (I) is not rst meaning; it is ``zero'' meaning.
The same conception of meaning applies to metaphor, and we should
understand how metaphorical use ``depends'' on literal meaning in exactly
the same way. The dependence should not be taken to mean that the lit-
eral meaning of the words determines their metaphorical use. As we saw
in examples (1)(10) in chapter 1, one expression with one literal meaning
(e.g., `is the sun') can be used metaphorically to dierent eects, or to
express dierent interpretations, in dierent contexts. But the literal
meaning of the word constrains its metaphorical eects or interpretations,
as I shall illustrate immediately.41 This role of the meanings of the words
used metaphorically is not their role insofar as they contribute consti-
tuents to an object of (literal) understanding of the utterance, to its literal
truth-conditions (where it has them). We don't need to understand how
the sentence uttered metaphorically would have been understood literally
in order to grasp the metaphorical use or interpretation of the sentence.
For the same reason, we do not need to know the literal truth-conditions
of the sentence interpreted metaphorically, knowledge of which, accord-
ing to Davidson, constitutes the understanding that, when shared, makes
for communication. Yet, the metaphorical use or interpretation of an ex-
pression F in a sentence . . . F . . . does depend on the literal meaning of F
(and, for that matter, of other individual words in the sentence) insofar as
our knowledge of the latter constrains our knowledge of the former. Thus,
as we shift our attention away from truth-conditions to constraints, both
for demonstratives and metaphors, we also shift away from Davidson's
conception of the domain for a theory of meaning. To illustrate the other
conception that focuses on the autonomy of meaning, another theme that
it is dicult to mesh with Davidson's conception, let me give one more
example of a meaning constraint.
Consider the meaning carried by a grammatical conguration, for
example,
(5) He saw John in the mirror,
From Metaphorical Use to Metaphorical Meaning 65

which cannot be understood as meaning

(6) John saw himself in the mirror.
The explanation of this fact appeals to a structural condition that pre-
vents the anaphoric pronoun in (5) from taking the noun (`John') as its
antecedent.42 The condition can be dened (among other equivalent
ways) in terms of an abstract ``tree'' structure that displays the grammati-
cal categories and dominance-relations among the elements in the con-
crete string. Given a tree structure for the string, let the scope of an
element be the least subtree to which it belongs. In simple subject-
predicate sentences, then, predicates are within the scope of subjects (since
the least subtree to which the subject belongs is dominated by a Sentence
node that also dominates the predicate and its arguments), but subjects
are not within the scope of arguments of predicates (since the least subtree
to which predicate arguments belong is [usually] dominated by a Verb
Phrase node that does not dominate the sentence subject). Given this ab-
stract notion of scope, we can then state the relevant condition on pro-
nomial anaphora:
(A) No element can seek its antecedent within its own scope.
Therefore, in (5), where the antecedent noun (a predicate argument) lies
within the scope of the pronoun (the subject), (A) is violated, ruling out
the possible interpretation. In (6), the subject noun lies outside the scope
of the reexive pronoun `himself ', allowing the interpretation that links it
to `John' as its antecedent.
Two features of this explanation are relevant to our present concerns.
First, what (A) codies is a kind of meaning carried by the conguration
or syntax of the sentence because it carries denite implications for its
truth-value. Second, as with demonstratives, this kind of meaning con-
strains the interpretation of the anaphoric construction without deter-
mining it. The condition does not state what the anaphoric interpretation
is or must be but what it cannot be; it excludes interpretations. Thus this
notion of meaning, like that of demonstratives, does not enter into the
content of what speakers understand by an utterance, but at most cir-
cumscribes the bounds of that understanding by ruling out impossible
interpretations.
Now, how, we might ask, would Davidson explain the impossibility of
understanding (5) as (6)? Recall his argument that because a theory of
meaning (even when formally reconstructed as a Tarski-style theory
of truth) is a highly context-specic theory of interpretation of utterances
66 Chapter 2

as rational communicative acts, it is nothing like a language (or grammar)

``if a language is anything like what many philosophers and linguists have
supposed'' (NDE 446), that is, a system of ``learned conventions or reg-
ularities'' (or an innate or innately xed body of principles or rules)
available in advance and independently of the occasion of utterance. In-
stead Davidson puts forth his picture of passing theories of interpretation,
which evolve from prior theories yet constantly undergo revision in the
course of language use in context. On this picture, one explanation of why
it is impossible to understand (5) as (6) might follow the principle that
just as we usually do not intend or try to do things we believe nature will
prevent us from doing, so we intend to say only what we believe our
interpreters will understand or only what we expect others will reasonably
be able to interpret us as saying.43 Hence, given our previous experience
and prior theory of our interpreters, we never would intend to mean (6)
by (5).
The diculty in applying this kind of Davidsonian explanation to our
data (such as (5)(6)) is that rules like (A) that exemplify the ``autonomy''
of meaning are, only very implausibly, the kinds of rules we would expect
to nd in passing theories. (A) applies to all anaphoric congurations and
can be formulated only by using syntactic categories and notions like
scope and dominance that go far beyond the properties of concrete utter-
ances. Likewise, (I) is an instance of a schema that holds for all indexicals
in every context in which they are used and rests on the abstract relation
of scope. For its notion of direct reference marks the fact that the inter-
pretation (or content) of the indexical is always xed relative to its context
of utterance, regardless of its scope in the sentence and regardless of
whether the matrix sentence also contains another expression, or opera-
tor, that makes its truth-value (as opposed to content) dependent on cir-
cumstances other than that of its context of utterance. For example, my
utterance of
(7) I might be assassinated
is true in my context of utterance if and only if there is some possible
circumstance in which I, JS, am assassinated; it is not enough for (7) to be
true that there be some possible circumstance in which whoever might be
speaking in that circumstance is assassinated. Thus (I) also expresses a
condition that applies not to the concrete utterance but to a signicantly
more abstract representation. Both (I) and (A), then, are more general,
xed, and abstract than the sort of context-specic expectations and
beliefs that prima facie govern the assignment of rst meanings according
From Metaphorical Use to Metaphorical Meaning 67

to Davidson's picture of prior and passing theories. This is not to say that
it would be impossible to formulate these conditions in Davidson's use-
oriented theory of interpretation. But any Davidson-style explanation of
the process whereby speakers come to learn these kinds of conditions,
appealing to mutual beliefs and expectations, would necessarily require
the attribution of rules of corresponding generality, systematicity, and
abstractness. And the more general, abstract, and xed these rules are, the
less will Davidson's account of passing theories be distinguishable from
the knowledge of language he rejects.
Furthermore, T-sentences (and the axioms of the truth-theory that en-
tail the T-sentences) known by an interpreter that give the meaning in the
form of truth-conditions for utterances that occur in actual use are plau-
sible constituents in prior and passing theories. But rules like (A) and
(I), as we've seen, do not themselves determine the content, or truth-
condition, of any utterance in any particular context; they only constrain
the possible interpretations. Hence it is not clear why they should be
included in a prior or passing theory keyed to the actual interpretations of
utterances in context.44
What lesson should we draw from this story? Davidson argues from the
premise that a theory of meaning is a theory of speakers' and interpreters'
shared understanding of communicative utterances to the conclusion that
our knowledge of such a theory cannot be knowledge of a language or
a grammar. But one philosopher's modus ponens is another's modus
tollens. The same argument could equally well be taken to show that if a
speaker's linguistic competence does consist in knowledge of a language,
or grammar, then a theory of that competence will not be a theory of
understanding, that is, a theory of interpretation (like Davidson's passing
theories) that yields truth-conditions of utterances in their respective con-
texts. This is not to deny that understanding an utterance (in one of its
senses) is knowing its truth-conditions. What is at issue is the assumption
that a theory of linguistic, and specically semantic, competence should
directly and fully explain the understanding of utterances. Linguistic
competence proper may instead be only one factor that contributes to
such understanding, and a theory of a speaker's knowledge of language
may be a theory of only one kind of knowledge specic to the linguistic
properties of utterances that contributes to understanding.45
The same argument applies to metaphor. A theory of a speaker's
semantic competence in metaphor, a theory of metaphorical meaning,
should not be assumed to be a theory of the complex ability to use or
communicate with metaphors, or a theory that directly explains what
68 Chapter 2

an interpreter understands by a metaphora theory that would deter-

mine for each utterance of a metaphor which feature or resemblance
it expresses on that occasion. Rather, from among the various com-
petencies, skills, and faculties that conjointly account for this complex
ability (including the interpreter's extralinguistic knowledge of the rele-
vant contextual parameter), a semantic theory of metaphorical interpre-
tation should address only the interpreter's knowledge of meaning specic
to metaphorical interpretation. The evidence for such a theory, to be sure,
will largely consist in actual utterances with their ``full'' metaphorical
interpretations in their respective contexts, but the semantic theory should
not account for that evidence per sethe interpreted utterances in their
respective contexts. The semantics should account only for the knowledge
that underlies the speaker's ability to assign interpretations that is specic
to their linguistic, or semantic, properties. The semantic theory of meta-
phor will consequently leave untouched a variety of aspects of metaphor
that depend on a larger model of speech performance or use: criteria of
appropriateness, the psychological processing of metaphors, and rhetori-
cal and noncognitive eects of metaphors that bear on their success.46 All
this and more: for if the ability to interpret a metaphor even in the limited
sense of assigning it a propositional content, or truth-condition, in a con-
text involves the contributions of multiple extralinguistic abilities, skills,
and presuppositionsincluding the ability to judge similarities or recog-
nize salient featuresthen the semantic theory will never suce to deter-
mine even the interpretation, truth-condition or propositional content, of
a single metaphorical utterance. An actual metaphorical interpretation of
an utterance in a context is underdetermined not only by the literal
meaning of the utterance but also by its metaphorical meaning.47
What work, then, does metaphorical meaning perform? And with what
aspects of metaphorical interpretation should a theory of a speaker's spe-
cically semantic competence in metaphor be concerned? To factor out
this component, we should begin with those facts about the interpretation
of a metaphor that would prima facie be a function specically of the
general idea of meaning. Meaning, we have proposed, provides con-
straints on what it is possible for speakers to intend to say with particular
types of expression. For evidence of metaphorical meaning, we should
therefore look for evidence of formal constraints on possible metaphorical
interpretations. As with demonstratives and our earlier example of ana-
phora, such a notion of metaphorical meaning is also autonomous of the
speaker's intentions; it rather serves to constrain how those intentions
say, the features of likeness we are ``made'' to see by the utteranceenter
From Metaphorical Use to Metaphorical Meaning 69

into the content of the interpretation. These autonomous aspects of met-

aphorical interpretation are the kind of data for which we should posit
metaphorical meaning as an explanation.
I'll conclude this section with two examples of constraints on the
potentially unlimited number of possible interpretations that one meta-
phorical expression (type) can be assigned in dierent contexts. These
constraints call out for metaphorical meaning.
The rst example involves a congurational constraint on anaphora
like our earlier data explained by (A). Let me begin with a nonmeta-
phorical illustration. It has frequently been observed that in cases of verb
phrase anaphora in which the antecedent allows multiple interpreta-
tions, the anaphor must always be given the same interpretation as the
antecedent. For example, `may' can be given the sense either of permis-
sion or possibility in
(8) John may leave tomorrow.
But in (9) and (10)
(9) John may leave tomorrow, and the same is true of Harry
(10) John may leave tomorrow, and Harry, too
the (italicized) pair of antecedent and anaphor in each string must be
given the same interpretation, either possibility or permission. This seems
to be a clear instance of the autonomy of meaning: The speaker's com-
municative intentions are constrained by a structural condition on the
``copying'' of the interpretation of the antecedent verb phrase onto the
anaphor.48 And because this constraint has obvious consequences for
the truth-value of the sentence, it is not simply syntactic but rather a
matter of semantic signicance, or meaning. Following the same kind of
reasoning I rehearsed earlier, it is also implausible to explain this condi-
tion on verb phrase anaphora as a condition on use, or in terms of an
interpreter's intentions or expectations a la Davidson. Despite my inten-
tions, I simply cannot mean by (9) or (10) that John has permission to
leave tomorrow, and it is possible for Harry to leave then. Sentences (9)
(10) are each only two-ways ambiguous rather than the four-ways we
would expect if it were only a matter of intentions and mutual expecta-
tions. Hence this is a constraint on interpretation that would appear to
call for semantic structure, that is, meaning.
Similar facts obtain for metaphor. Contrast the literal interpretation
(i.e., content or truth-condition) of `is the sun' in the sentence (repeated
here)
70 Chapter 2

(11) The largest blob of gases in the solar system is the sun
with its dierent metaphorical interpretations in
(12) Achilles is the sun
and
(13) Juliet is the sun.
As I said in chapter 1, `is the sun' varies in its metaphorical interpretations
in these strings, and with some systematicity, as a function of some dif-
ference in its respective context, either linguistic or extralinguistic (or
both). Consider now the following (semantically) ill-formed sentences,
which are examples, again, of verb phrase anaphora:
(14) *The largest blob of gases in the solar system is the sun, and Juliet/
Achilles is, too,
in which the rst conjunct has the same interpretation as (11) and the
second, the interpretation of (13)/(12); and
(15) */?Juliet is the sun, and Achilles is, too,
in which the rst conjunct has the same interpretation as (13) and the
second, the interpretation of (12).
Each of these ill-formed strings would seem to be the result of violating
the same kind of interpretive constraint at work in (9)(10). That is, if
the interpretation of the antecedent is copied onto the anaphor, both the
antecedent and the anaphor must have the same interpretation (whatever
it is). Hence in (14) where the interpretation of the antecedent is the literal
meaning of `is the sun' and the interpretation of the anaphor would seem
to be metaphorical (on pain of absurdity), we have one violation. In (15)
(which is somewhat more acceptable to informants), we have two dier-
ent metaphorical interpretations of `is the sun', also violating the con-
straint.49 Both interpretations are ill formedalthough, as is often the
case with such gures, we try to impose an interpretation on the strings
despite the violation. However, it is precisely the feeling of play or pun
that accompanies such imposed interpretations that gives away the
underlying semantical ill-formedness of the strings.
These are examples in which certain aspects of metaphorical interpre-
tation show themselves to be autonomous ofprecisely because they
serve as constraints oninterpreters' intentions. As with the earlier
examples, it is dicult to see how we might account for these constraints
in terms of use or mutual beliefs and expectations. What is needed is a
From Metaphorical Use to Metaphorical Meaning 71

structural conditionthe same condition that applies to verb phrase

anaphora in generalin which case we must attribute to metaphor the
semantic structure, or meaning, necessary for the requisite condition to
apply. Hence metaphorical meaning.
A second example of a constraint on metaphorical interpretation is
analogous to the constraint on demonstrative content specied in (I) and
exemplied by our earlier discussion of (7). Even when a demonstrative
occurs within the surface scope of a modal operator, its interpretation is
determined with respect to its context of utterance and never with respect
to the circumstance with respect to which the truth-value of the sentence is
determined. Similarly, the interpretation, that is, the truth-condition or
content, of a metaphor, which systematically varies with its context (as we
saw in ch. 1), is always xed relative to its actual context of utterance,
never relative to any other context or circumstance. Even if the metaphor
occurs within the scope of a modal operator, it is never interpreted rela-
tive to the counterfactual (or alternative possible) circumstance(s) in
which we evaluate whether it is true. Suppose that Paris disagrees with
Romeo's utterance of `Juliet is the sun' but concedes that
(16) Juliet might have been the sun.
(16) will be true in (the circumstance of ) its context (call that w(c)) just in
case there is some possible circumstance w* in which Juliet has the par-
ticular set of properties P which is the content of `is the sun' interpreted
metaphorically in its context c. That is, Juliet must fall in the extension at
w* of the relevant property P, but the relevant property P is xed relative
to the context of utterance c, not relative to the counterfactual circum-
stance w*. It is not sucient for (or relevant to) the truth of (16) that
Juliet possesses whatever property would be the content of `is the sun'
were it interpreted metaphorically in w*. Now, as in the case of (I) for `I',
the kind of condition that accounts for this fact does not determine the
interpretation of the metaphor; it only species what the interpretation
cannot be. That is, it only constrains the interpretation. And again, it is
dicult to see how to explain the presence of this constraint in terms of
use, intentions, or mutual expectations. The constraint calls for meta-
phorical meaning.50

VI Is a Semantic Theory of Metaphor Possible?

Thus far I have argued that if we try to explain metaphor as a use of

languagestarting from its high degree of context-dependencewe
72 Chapter 2

nonetheless need a notion of metaphorical meaning. But not just any

notion. Davidson is right to criticize previous theorists of metaphor for
playing fast and loose in their talk of meaning. If a notion of metaphori-
cal meaning must meet the same conditions of explanatory adequacy we
require for nonmetaphorical meanings, the particular feature the meta-
phor makes us see on an occasiona novel aspect or similaritycannot
be a meaning of the metaphor. On the other hand, if we require (as
Davidson also does) that any account of metaphor must explain how the
metaphorical depends on the literal, Davidson's own explanationthe
causal dependence of metaphorical use on literal meaningfalls short at
crucial moments, leaving gaps that can be lled only by semantic notions
like meaning that spell out constraints on what we can metaphorically use
words to express. The same kinds of constraints, I have also argued,
constitute the meanings (as distinct from truth-conditional content) of
classic context-dependent expressions such as demonstratives. But once
Davidson opens his own semantic theory to admit demonstratives (as he
must), if we can show the context-dependence of metaphors to be no dif-
ferent in kind from that of demonstratives, he should welcome metaphors
as well.
It is one thing, however, to show that objections do not rule out all
possible notions of metaphorical meaning or that there exists a domain of
phenomenathat is, the kinds of constraints that call for meaningthat
ought to fall under a semantic theory. It is another matter to show that,
and how, we can actually construct a theory for metaphor that would
satisfy the requirements for a nitistic semantics of the same kind that we
have successfully begun to construct for signicant fragments of natural
language. Davidson denies this possibility (thesis 4 above), and prima
facie he has good reason to do so. A nitistic semantic theory explains
how the nite speaker knows the interpretations of an innite set of sen-
tences (or complex expressions) by showing how, in virtue of their struc-
ture, each sentential (or complex) interpretation is the product of a nite
number of applications of a nite number of recursive rules operating on
a nite number of semantic primitives (e.g., the truth or satisfaction con-
ditions of a nite lexical vocabulary). If metaphorical interpretations
could be explained by a nitistic theory, they would have to be either
interpretations of the nite number of semantic primitives or the products
of rules of composition. But we run into major obstacles when we try to
pursue either of these routes.
In section III, we reviewed various diculties that arise if we take
a metaphorical use or interpretation to depend on, or be a function of,
From Metaphorical Use to Metaphorical Meaning 73

either its literal extension or its literal intension; the same arguments leave
the compositionality of metaphor at best an open question. Furthermore,
we noted in chapter 1 (examples (1)(10)) how the metaphorical content
of any individual expression apparently depends on features of its larger
congurationof at least its containing sentence but, as we shall see, also
its wider extralinguistic context. This reverses the standard ``bottom-up''
order of compositional interpretation that constructs the meanings of
compound expressions from the isolable meanings of their parts by strict
rule-by-rule combination. These considerations make it doubtful that
metaphor falls on the compositional side of semantics.
Nor do the phenomena of metaphor constitute a case of meaning of the
type that we would identify with a semantic primitive to be explained by
the lexical branch of semantics. Despite some important advocates such
as Monroe Beardsley (1978), who claims that ``there are indeed meta-
phorical senses'' that ``behave in many of the same ways as literal senses''
(11)and his evidence includes fallacious inferences like our earlier ex-
ample (1)(3)the idea that a metaphorical interpretation is just another
primitive word-sense cannot be the whole story.
First, as we have observed, metaphorical interpretations of one expres-
sion (type) vary with the contexts in which it is tokened. But there is no
antecedently xed upper bound on the number of dierent contexts in
which a given metaphorical expression (type) can be tokened; hence there
is no antecedently xed upper bound on the number of distinct meta-
phorical interpretations of the same expression (type) that speakers are
able to produce and comprehend. This leads to a familiar predicament.
If each of its metaphorical interpretations is simply a distinct, primitive
sense, the ability to interpret a metaphor requires the mastery of an un-
bounded number of senses for each expression. No matter how many
metaphorical interpretations of a given expression a speaker has mastered
at a time, each new metaphorical interpretation in a signicantly dierent
context is another sense to be learned anewor miraculously intuited. An
account in terms of ambiguity simpliciter that does not spell out how the
dierent interpretations of one expression are systematically related
hardly describes this feat.51
Second, if metaphorical interpretations are just additional senses, then
expressions with metaphorical and literal interpretations are simply am-
biguous. But metaphor does not t neatly into the standard classication
of ambiguities, which suggests that a metaphorical interpretation is not
simply an additional sense of an ambiguous expression. Consider the
idiosyncratic ambiguity exemplied by words like `ear' (used of corn and
74 Chapter 2

of the bodily organ) or `corn' (used of the vegetable and the growth on the
foot).52 Their dierent senses are mutually independent in that knowledge
of one does not require knowledge of the other, whereas knowledge of the
metaphorical interpretation(s) of an expression does require knowledge
of its literal interpretation (in whatever exact way). As Davidson also
notes, if we leave out all such ``appeal to the original [or literal] meanings
of the words . . . all sense of metaphor evaporates'' (WMM, 248249).
But if we fully assimilate metaphor to idiosyncratic lexical ambiguity, we
lose this dependence.
On the other hand, expressions with metaphorical and literal interpreta-
tions are also unlike examples of the standard type of systematic lexical
ambiguity, for example, `book', which has abstract and concrete senses in
(17) and (18), respectively:
(17) John wrote a book.
(18) The book weighs ve pounds.
One dierence between the metaphorical/literal case and systematic
lexical ambiguity emerges in that gray area where syntax and semantics
interact: although the explanations for their dierent types of behavior
are not well understood, the systematically lexically ambiguous expres-
sions appear to undergo syntactic transformations that expressions with
metaphorical and literal senses do not. The two dierent senses of `book',
for example, can form a relative clause, as in
(19) John wrote a book that (which) weighs ve pounds53
unlike the expression `is the sun' in
(20) *Juliet is the sun, which (that) has a diameter of roughly 865,000
miles
which is metaphorical as the verb phrase of the main clause and literal
as the antecedent of the relative pronoun. Similar resistance to relativiza-
tion is found where the expression has two dierent metaphorical inter-
pretations, as in
(21) *Juliet is the sun, whose burning fury consumes the life of Troy.
Dierences of this kind, little understood as they are, indicate that, if
metaphor is an instance of systematic ambiguity, it is not of the standard
lexical type.
In sum, metaphor falls neatly under neither compositional (sentence)
nor lexical (word) semantics. Since these two exemplify the main branches
From Metaphorical Use to Metaphorical Meaning 75

of contemporary semantics, one conclusion one might draw is that meta-

phor lies entirely outside semantics. Add to this the context-sensitivity of
metaphorical interpretationtogether with the assumption that seman-
tics is task-specic and metaphor depends on all sorts of non-language-
specic skillsand it is easy to appreciate why many philosophers and
linguists conclude that metaphor belongs under theories of use, speech-
acts, or pragmatics.
In response, our main task is to show how a nitistic, compositional se-
mantic theory of metaphor is nonetheless possible. In chapter 3 I embark
on this positive project by lling in the necessary background adapted
from the semantics of demonstratives; in chapter 4 I begin to apply the
apparatus to metaphor.
Chapter 3
Themes from Demonstratives

Two themes from the theory of demonstratives feature in my account of

metaphor. The rst is the distinction between the propositional content
(for short, content) and the character of an expression. The second is a
generalization of the idea of a demonstrative from a handful of expres-
sionsthe pure demonstratives (`This', `that', `she', `then', `thus', etc.)
and proper indexicals (`I', `here', `now', `actual', today', etc.)to a much
broader class of expressions whose interpretation in a context (or use on an
occasion) is demonstrative. (As noted in ch. 2, I use the unqualied
`demonstratives' as a generic term for all of these expressions.) Both of
these themes are directly adopted from David Kaplan's seminal work on
demonstratives. But my aim here is not Kaplan exegesis, and so my ex-
position will anticipate the application of his work to metaphor.1
Two preliminary terminological remarks: One obvious dierence be-
tween demonstratives and indexicals is that they depend on dierent
features of their respective extralinguistic contexts for their referents.
However, Kaplan has also noted a deeper dierence. Knowledge of the
linguistic meaning of an indexical (type) is both necessary and sucient to
determine the direct referent of a token; any accompanying nonlinguistic
act like a pointing is either redundant or overruled. In contrast, knowl-
edge of the linguistic meaning of a demonstrative (type) is never sucient
to determine its referent (on a tokening or utterance). Its linguistic mean-
ing must be completed by an additional nonlinguistic factor, a demon-
stration. That is, a demonstrative (type) without a demonstration (that
completes a tokening) falls short of being a semantically signicant unit; it
is an incomplete expression. To be sure, there are nondemonstrative
uses of the word `that' that lack demonstrationsfor example, `there was
that book by Kant on pure reason'but for a use of `that' to count as
demonstrative I shall assume, following Kaplan, that it is necessary that
(a tokening of ) it be completed by an extralinguistic demonstration.
78 Chapter 3

Furthermore, the completing demonstration is necessary both where the

demonstrative is simple and where it is linguistically modied, either as a
complex demonstrative such as `that balding philosopher with the limp'
or when it contains a lexically built-in sortal like `she' ( `that female').2
When completed by its respective extralinguistic demonstration, I refer to
the mixed linguistic-extralinguistic entity, that is, the type of the token, as
a complete demonstrative.
Second, to anticipate our discussion in chapter 4, note that metaphors
are a hybrid: In some respects, they are like the pure indexicals and in
others, like Kaplan's dthat-descriptions, that is, descriptions interpreted
(or used) demonstratively. The expressions they are least like semantically
are the complete demonstratives, although the idea of a dthat-description
is derived from Kaplan's analysis of complete demonstratives. My dis-
cussion will reect this unbalanced set of similarities and dierences,
giving the most attention to indexicals and dthat-descriptions.

I Some Prehistory

Kaplan's distinction between character and content emerged in the 1970s

as part of a critique by proponents of the ``New'' theory of reference of
the then-reigning doctrine of reference: ``Frege's theory,'' ``Fregeanism,''
the ``Frege-Russell theory,'' ``Descriptionalism,'' or, as I'll simply call it,
the ``Old'' theory of reference.3 The basic idea of the Old theory was that
associated with each expression in the language, there is a fully con-
ceptualized, purely qualitative representation that individuates the object
that is its referent or extension. This representation plays at least ve roles
in the theory: (1) It furnishes the conditions to be satised in an (idealized)
mechanism or procedure by which the extension or referent of the ex-
pression is xed, that is, the factor contributed by the expression to the
truth-value of its containing sentence. (2) It is what a speaker knows when
he understands the expression. (3) It is, or expresses, the linguistic mean-
ing of the expression. (4) It contains a mode of presentation, or way of
thinking, of the referent that accounts for the ``cognitive signicance'' or
``information value'' of the expression; as such, it is employed to solve
problems like Frege's puzzle of identity. (5) It is the object of psycho-
logical attitudes and indirect discourse: what is believed, said, or asserted.
Finally, on the Old theory, one and only one thing performs all ve of
these functions.
The paradigmatic type of expression according to the Old theory
is the singular term and, among singular terms, the complete denite
description `The F1 & . . . & Fn '. Since its conjunction of predicates
Themes from Demonstratives 79

F1 & . . . & Fn explicitly states the set of descriptive conditions the unique
satisfaction of which is the mechanism by which its referent is xed, the
complete denite description transparently carries its associated repre-
sentation on its sleeve. Let's say that a singular term (or an occurrence of )
b denotes an object b if and only if there is a conceptual complex C
of individuating descriptive or qualitative conditions associated with b
(either relative to or independent of a context), and b uniquely satises C.
(If nothing satises the conditions, then the term denotes, and thereby
refers to, nothing.) This mechanism makes reference a matter of ``t,'' and
determination of the referent (at a circumstance) is conditional on, or
mediated by, the complex C. The reference relation itself is, then, indirect.
What the singular term directly contributes to the (truth) evaluation of its
sentences is C, and the referent is whatever turns out to satisfy C. Hence
one apparent implication of the Old theory is that in dierent circum-
stances of evaluation the same singular term, with the same associated
representation C, might denote, and refer to, dierent individuals or
extensions, namely, whatever happens to satisfy C in that circumstance.
That is, the Old theory seems to imply that, if we take the complete de-
nite description as the standard from which we generalize, singular terms
are, in now familiar terminology, nonrigid designators.
Since the 1970s, the Old theory has been subjected to a barrage of well-
known criticisms by New theorists for its conception of the rst role of the
associated representation. I shall not review all the objections, but the
New theorists have drawn two general morals.4 First, although it may be
the case that complete denite descriptions are nonrigid, New theorists
argue that this is not so for all singular terms. In particular, paradigmatic
singular terms, proper names and demonstratives (as used in a given
context), refer to the same individual or extension in all possible circum-
stances of evaluation; that is, they designate rigidly.5 Second, New theo-
rists challenge the Old theory's account of how we determine the referents
or extensions of proper names, demonstratives, and even some denite
descriptions. In place of satisfaction, they argue that a speaker is able to
refer successfully with many of her singular terms without possessing the
rich, fully conceptual, purely qualitative knowledge required for denota-
tion. For in addition to the (limited) conceptual resources she possesses,
she also employs contextual relations, extralinguistic heuristics, and her
connections to other members of the linguistic community from whom
she acquires her language. In sum: Not by concepts alone do we refer.
The Old theory has fared better with respect to some of the other roles
for its associated representation, especially the fourth and fth roles that
capture its epistemology. According to this general doctrine, which I'll
80 Chapter 3

call perspectivalism, all attitudes, knowledge, and referencenamely, all

thought and speech about objectsare from a particular perspective. Just
as we always perceive some thing from some denite angle or direction,
whenever we think about (refer to, talk about) a public object it is from
some specic point of view.6 Since nothing is ever seen in its ``totality,''
that is, simultaneously from all perspectives, each perspective might be
said to furnish a ``partial'' view of the referent. This partiality of a per-
spective is what enables the associated representation to do its cognitive
work: to solve Frege's identity puzzle and account for belief reports. Each
such partial perspective also corresponds to a (Fregean) mode of pre-
sentation, or way of thinking of a thing that is (conventionally or con-
textually) associated with a term. Just as we often fail to realize that an
object perceived from one perspective is the same object perceived from a
second perspective, so we fail to identify one object represented in two
ways, corresponding to two terms. Hence it is informative to learn the
truth of an identity statement anked by dierent terms (types), asso-
ciated with dierent representations, unlike one anked by the same term
(type) with one representation.
Kaplan's theory of demonstratives shares this perspectivalist epis-
temology, although unlike the classical Fregeans who account for it in
terms of sense (their associated multipurpose representation), he explains
it with his notion of character. In chapter 7, I'll also employ the idea of
metaphorical character to explain how metaphors contribute a distinctive
cognitive perspective on their contents. So, insofar as Kaplan and I take
the cognitive phenomena as ``data'' to be explained by a semantic theory,
there remains a deep Fregean current running through our accounts.7
However, the Old theory goes beyond this general form of perspectival-
ism in two ways. First, it holds not only that we refer from a perspective
or think of things in a particular way but also that these perspectives or
ways of thinking of things are always expressed, contained, or encoded in
purely conceptual or qualitative, context-transcendent representations.8
Second, it holds not only that all thought or speech about an object is
from a perspective but that the perspective determines the object we are
speaking or thinking about. Therefore, the conceptualized representation
of the object must be as specic and complete as denotation requires.
Kaplan and I reject these two further theses together with the thesis that
denotation is the mechanism that xes reference. To account for the cog-
nitive phenomena, it is enough if there is some dierence between the
conceptual perspectives ``contained'' in the representations (or characters)
associated with dierent co-referring singular termseven if neither per-
Themes from Demonstratives 81

spective is complete or purely conceptual. This, I will argue, suces both

for demonstratives and metaphors.

II Character and Content

Much of the critique of the Old theory has centered on proper names. But
demonstratives and indexicals also play a prominent partwith slightly
dierent morals. Consider, again, the rst person indexical `I'. Any ac-
count of this expression must explain, to begin with, (i) how its linguistic
meaning, the rule (I) (repeated here, with slight modication)
(I) Each occurrence of `I' directly refers to the agent of the context, e.g.,
its speaker
remains the same for all speakers and across all utterances, while (ii) the
referent of each utterance of `I' varies with its speaker. If (I) is (roughly)
the Old theory's associated representation for `I', these two data are
already in conictthe same representation should determine the same
referent on all utterances. But this is not all. The Old theory identies the
meaning of a term (the third role of the associated representation) with its
propositional or truth-conditional content (the rst and second roles). But
if you (Max) and I (JS) each utter
(1) I am laughing,
not only do the referents of our two utterances of `I' dier; because my
utterance might be true and yours false, what is said by our utterances,
their propositional contents or truth-conditions, must also dier. My
utterance of (1) generates the proposition represented as
(1a) hJS, Is-laughingi.
Yours:
(1b) hMax, Is-laughingi.
So, while its linguistic meaning remains constant, the propositional con-
tent or truth-condition of (1), and not only the referent of the indexical,
also varies from utterance to utterance. Meaning and propositional or
truth-conditional content also, then, cannot be identical, and at least two
substrata must be distinguished within the Old theory's multipurpose
associated representation. Following Kaplan, we shall call these its char-
acter and (propositional) content.
The content of a sentence (uttered in a context), namely, a proposition,
is what is evaluated at a circumstance to determine its truth-value (at that
82 Chapter 3

circumstance). The contents of terms, predicates, etc. are their contri-

butions (whatever they turn out to be) to their containing propositions.
Circumstances of evaluation can include any feature with respect to which
we may evaluate the truth-value of a sentence (or, more generally, the
extension of an expression), but I shall generally take them to be simply
worlds. In a possible worlds semantics, contents can be represented as
intensions: functions from the set of possible circumstances to the appro-
priate type of extension. But for our purposes, no particular formal
analysis of content is necessary. I'll return in the next section to the sen-
tence-like structure of (1ab).
This notion of content is more or less familiar; the novel element in
Kaplan's proposal is character, the general notion he proposes to corre-
spond to the constant knowledge of speakers across utterances such as (I)
for `I'. Technically, character is the semantic property (or value) of an
expression (type) that determines, for each context of utterance, the con-
tent of (a token of ) the expression in that context. As in (I), characters
can be stated informally as rules that specify the features of the context on
which their content depends. But like contents, Kaplan also proposes that
we represent characters as functionsfrom the set of possible contexts to
the appropriate type of content.9 All expressions in the language, it
should be emphasized, are assigned characters, but only those expressions
that are context-sensitive have nonconstant characters: characters that
determine dierent contents in dierent contexts. Eternal expressions have
constant characters: They determine the same content in every context.10
Unlike the Old theory, which assigns two semantic values to each
expression (sense/meaning/content and reference), we now assign three:
reference, content, and character. Analogously, we can distinguish three
stages, at each of which we raise a dierent question about an utterance
and at each of which context functions in one of the three roles we dis-
tinguished in chapter 2.
Suppose a sound event occurs. The rst question is whether the event is
an utterance of a well-formed expression (a word with a particular
meaning) in a language and, if so, which one. In answer to this question
we assign the event a linguistic description that includes a rule of charac-
ter that determines how the event, identied as a token of a certain type in
a certain language, should be linguistically understood on that occasion.
For example, to the sound ` ' we assign the character of the rst person
indexical `I'. Call this the character-assignment stage; as I said in chapter
2, context informs this assignment, but its role here is presemantic.
Themes from Demonstratives 83

With the utterance assigned a character, its occurrence in some context

generates a content for the expression in that context. Call this the con-
tent-generation stage in the interpretation of the utterance.11 At this stage,
we ask: Given the expression type assigned to the utterance, what is being
said? What factor does the utterance of `I' contribute to those that must
be taken into account in evaluating (at the next stage) whether the sen-
tence in which it occurs is true or false? At this stage, context again plays
a central role. Thus, for `I', corresponding to the character (I) it was
assigned, the context furnishes an individual, namely, the agent of the
context or speaker. This role of the context is semantic: Part of a speaker's
semantic competence consists in his knowing, for every expression, the
appropriate extralinguistic parameter (if any) in any context that deter-
mines what its content would be (or what the utterance would say) in that
context.
Finally, given the content of the utterance that was determined at the
previous stage, its evaluation at a circumstance yields a truth-value for the
sentence uttered. Call this the truth-evaluation stage. Here we ask: Is what
is said, the content of the utterance, true or false? Again context comes
into play because we are typically concerned with the actual truth-value of
the utterance, that is, its truth-value in its actual circumstancesthe cir-
cumstances of its context. But this role of the context is postsemantic, for
the truth-value of the utterance is signicant only insofar as the utterance
is used to make a particular kind of speech act, an assertion (or some
other act or use that presupposes assertion). If the sentence were to be
uttered (also) for some nonassertoric purpose, context would, most likely,
again enterbut for an evaluation of a dierent kind of appropriateness.
In either case, the context is functioning postsemantically.

III Direct Reference: Singular and Predicative

I said in section I that the New theory critics of the Old theory of refer-
ence challenge both its implication that all singular terms (hence demon-
stratives) are nonrigid designators and its denotational model of the
xation of reference. In contrast, Kaplan claims that demonstratives
(along with proper names and certain common nouns such as names of
natural kinds) are not only rigid but also directly referential terms. Again,
I won't rehearse his arguments for this thesis. However, I do want to
emphasize what the claim means, rst for singular demonstratives
which have been Kaplan's almost exclusive focusand then predicate
demonstratives. This is important for our project because metaphors, as
84 Chapter 3

I'll argue in chapters 4 and 6, are primarily predicative but are also
directly referential.
First, the thesis that a demonstrative is directly referential isbeyond
the claim that it is rigid, which is a thesis about its referent or extension
a thesis about how its referent or extension (at a circumstance) is deter-
mined.12 And since content is what determines reference (at a circum-
stance), it is a thesis about, and exclusively about, the content of the
demonstrative. Second, the thesis is best understood by the via negativa:
What makes a term directly referential is that it is not indirectly referential
and, in particular, that it is not denotational. The content, or proposi-
tional contribution, of a directly referential term is not a qualitative, con-
ceptual individuating condition by which it denotes its referent (at a
circumstance). It is not even a concept or qualitative condition that turns
out to denote the same individual at all circumstances, a consequence that
would be compatible with the term's being rigid. Kaplan depicts this
nondenotational nature of direct reference by incorporating the very in-
dividual who is the (rigidly designated) referent into its respective propo-
sition. This depiction, as we'll see next, is also potentially misleading
but the force of the image is its exclusion from the propositional content
of a conceptual complex in virtue of which the term could be said to de-
note its (constant) referent.
In saying that the direct reference thesis is a claim only about the con-
tent of the demonstrative, I also intend to emphasize that the thesis says
nothing, pro or con, about additional kinds of ``meaning'' or ``linguistic
signicance'' other than content that the expression might possess. This
should be underlined for two reasons: (1) Kaplan's image of an individual
as propositional content may reinforce the opposite impression; and (2)
Kaplan's own paradigm of a directly referential term, the free variable
under an assignment, in fact has no additional meaning or signicance.
Everything that might be said about the way in which its value (which is
its content) is assigned, or generated, is presemantic: prior to the assign-
ment of any semantic value or signicance to the expression. Similarly for
logical constants, that is, variables with a constant assignment relative to
a language, and ordinary proper names whose referents are ``assigned''
through a historical (or causal) chain originating with a baptism or
dubbing, a story that is, most plausibly, also presemantic.13 For these
expressions, denotation plays no role at any stage, either when their con-
tent is evaluated at a circumstance or when the content is itself generated.
The only semantic value or signicance these expressions possess is their
referent, the element they contribute to the truth-valuation of their sen-
Themes from Demonstratives 85

tences. These expressions, I shall therefore say, are not only directly ref-
erential (in content) but also thoroughly nondenotationalto exclude de-
notation from any stage.
Kaplan formally represents the contents of sentences containing di-
rectly referential terms as sequences containing individuals, times, and
other (direct) referentswhat are now called Russellian or singular prop-
ositions; examples are (1ab). These quasi-sentential Russellian propo-
sitions are meant to capture aspects of logical syntax and semantic
structure that are lost in set-theoretic conceptions of a proposition, such
as the possible worlds model of a proposition as a function or (charac-
teristic) set of possible worlds.14 However, there are still other aspects of
logical structure not captured in representations like (1ab); for example,
complex singular demonstratives have complex logical structure, repre-
sented in their character, but simple content, an individual. To capture
this kind of logical but nonpropositional structure without abandoning
Russellian propositions, we can borrow some formal machinery from
Kaplan. If F in an expression, let {F} stand for its character. (I suppress
reference here to the model for our interpretation and an assignment
function for any free variables that occur in F.) We can then represent the
content of F in a context c as {F}(c). Thus my utterance of (1) in c can be
represented not only as the Russellian proposition shown in (1a) but also
as the Russellian proposition (1a 0 ):
(1a 0 ) h{I}(c), Is-laughingi.
Note that the propositional constituent represented in (1a 0 ) by `{I}(c)' is
the individual JS as in (1a); the character of `I' is not itself a propositional
constituent. Hence (1a 0 ) is a bonade Russellian proposition, indeed the
identical Russellian proposition as (1a), although, unlike it, (1a 0 ) displays
``character-istic information'' not preserved in its content. We'll see some
examples of the importance of this character-istic information both for
demonstratives and metaphors.
Kaplan calls propositions that contain only properties and logical
functions but no individuals general propositions. This Russellian charac-
terization reects the fact I mentioned earlier, that discussions of direct
reference have concentrated almost exclusively on singular terms. But
some New theorists have argued that certain common nouns also name
natural kinds, substances, and other abstract entities; this suggests that
these nouns should also be treated as directly referential termsbut
with general rather than individual referents. Closer to home, there are
demonstratives that occur in predicative position such as `is that F' (e.g.,
86 Chapter 3

shape, color, style, speed, etc.) and `thus' (as in `the such-and-such Vs
(looks, walks, smells, etc.) thus', where the predicate demonstrative is
completed by a demonstration of how the thing looks, walks, etc.).15
Should such propositions count as singular or general?
For present purposes, let's restrict our attention to predicate demon-
stratives. One issue is what to count as their referents. If their referents are
their extensions in the usual sense, namely, the sets of things at a circum-
stance of which the contents of the predicates are true (or to which a
predicate, as uttered in a context, applies), then (most) predicate demon-
stratives are not even rigid, let alone directly referential. Consider the
predicate demonstrative `is that shape' in the sentence `The coee table I
want to buy is that shape [speaker points at an oval-shaped mirror]'. The
predicate demonstrative `is that shape' has a nonconstant character:
depending on the shape demonstrated, it will have dierent contents in
dierent contexts (being oval, being square, etc.). But unlike the kinds or
substances named by natural kind or substance terms that are prima facie
essential (or necessary) to any object that falls under them in any one
circumstance, the property of having a certain shape is clearly contingent;
its extension will vary from circumstance to circumstance. Predicate
demonstratives, then, have nonconstant characters but their contents do
not determine the same extensions in all circumstances. Hence, if direct
reference requires the rigidity of its extension, predicate demonstratives
are not even rigid, let alone directly referential.16
There is another alternative, however. Instead of taking the referent of
a predicate as its extension (at a circumstance), let's think of a predicate
(or certain predicates) more like a singular term referring to a property.
(We can also assign the predicate an extension at each circumstance,
namely, the set of individuals at the circumstance who possess the prop-
erty.) My motivation for this is not metaphysical. I want to articulate the
dierence between the contents of predicatesthe factors they contribute
to truth-valuationsthat we captured with singular terms by distinguish-
ing direct from indirect reference, that is, by distinguishing between deter-
mining the reference by denotation, by satisfaction of a conceptualized
representation associated with the expression, and by other ways.17 Some
predicates express (refer to) properties in virtue of there being associated
with them a fully conceptualized representation (known by their speaker)
that the property satises; others express (refer to) their properties in some
not fully conceptualized, context-exploiting manner. A predicate demon-
strative is directly referential just in case it is not indirectly referential. The
mode of reference- or expression-determination that holds between the
Themes from Demonstratives 87

predicate demonstrative and its property is not mediated by a purely

conceptualized representation.18
To return now to the question of whether the proposition expressed by
a sentence containing a predicate demonstrative is singular or general:
Suppose, following Frege, that all expressions, predicates as well as terms,
have referents, and that the referents of predicates are properties. Let's
also distinguish, following Russell, between propositions whose con-
stituents include the very individuals, properties, and relations, on the
one hand, and those whose constituents are purely conceptualized, quali-
tative representations whose conditions determine individuals, properties,
and relations by denotation (or a suitably generalized relation) at cir-
cumstances. Let's call the latter purely conceptualized propositions; the
former, referential propositions. The Old theory holds that all propositions
are purely conceptualized; according to the New theory, at least some
propositions are referential. The class of referential propositions includes
Kaplan's singular propositions, but it also includes propositions whose
constituents are properties and relations ``themselves'' rather than con-
ceptualized representations that denote (or denotationally express) them.
As with the singular direct reference thesis, the distinction between
these two kinds of propositions is entirely at the level of content. The
``generalized'' direct reference thesis holds that, for at least some predi-
cates or common nouns, we should distinguish the property to which it
refers from the means by which we conceive of, or express or refer to, that
property. In some cases, we want to know whether something falls under
a particular property (at a circumstance) regardless of how that property
is conceived or expressed; in other cases, our concern is whether some-
thing falls under the property expressed or represented by some predicate
(or other conceptual representation), whatever particular property that
turns out to be. This distinction also reects our referential skills and
capacities. Sometimes we refer to the property by means of a purely
qualitative representation we possess in our conceptual repertoire. In
other cases, we take advantage of our contextual relations to the property,
using a sample we display or demonstrate. And sometimes this is because
we do not know, or because there isn't, a label or concept for that prop-
erty to enable us to refer to or express it in a context-independent mode.
A similar situation, I shall argue in chapter 5, holds for certain meta-
phors whose context-dependence generates referential propositions. These
metaphors, like predicate demonstratives, directly refer or express prop-
erties in the absence of, or without employing, fully conceptualized rep-
resentations of the properties. Through this means, we can explain how
88 Chapter 3

metaphors enable us to express novel properties, that is, properties for

which we possess no well-integrated conceptualization in our linguistic
and cognitive repertoire.

IV Indexicals and the Parametric Determination of Their Referents

I argued in section III that the direct reference thesis for demonstratives
consists in (i) the claim that they rigidly designate their referents or
extensions and (ii) the claim that the determination of their referents (at
circumstances) is not indirectly referential, that is, by use of a mechanism
like denotation. This leaves us with the question: If not by denotation,
then by what kind of mechanism do we determine the (direct) referent
(which, for Kaplan, simply is the content) of a demonstrative? For varia-
bles, constants, and proper names, we relegated the assignment of their
values, or direct referents, to presemantics. What I'll argue in the re-
mainder of this chapter is that there are signicantly dierent answers
to this question for the two kinds of demonstratives that will serve (in
ch. 4) as the models for our hybrid analysis of metaphor: the proper
indexicals and a new species of demonstrative invented by Kaplan, the
dthat-description (or demonstratively interpreted description). (Complete
demonstratives are yet another, still more complicated story; for reasons
of space I shall discuss them only insofar as it is necessary to present
dthat-descriptions.) In this section, I'll begin with the indexicals.
The direct referents of indexicals would seem to be determined (for
each context) by their rules of character, such as (I) for `I'. But exactly
how do these rules work? On the face of it, the way that, say, the rule of
character (I) for `I' functions is not very dierent from the way that the
Old theorist's associated representation C operates according to its rst
three roles: It is the linguistic meaning of the word that a speaker under-
stands and it xes its referent denotationally. In other words, (I) is
something any competent speaker of the language is presumed to know as
part of knowing the linguistic meaning of `I', and its rule of character is
semantically signicant information that the indexical carries beyond its
semantically signicant direct referent (content). There are at least three
reasons for holding this. (1) As I argued in chapter 2, a primary function
of linguistic meaning is to constrain the intentions we can express with a
particular linguistic form on any occasion of use. The character rule (I)
spells out exactly what we would expect of the constraints on `I'. (2) There
is a logic specic to the indexicals: There are inferences and sentences
(e.g., `I am here now' and `I exist') whose ``validity'' (i.e., truth in every
Themes from Demonstratives 89

possible context in which they are uttered) depends on their (indexical)

characters and not on their contents in any particular context (Kaplan
[1989a], 541553). (3) We know that sentences like `I am here now' are
true, or express a truth, in the circumstance of every context in which they
are uttered even though we do not always know what truth is expressed by
each utterance in its context. The explanation for this dierence is that we
know the former solely in virtue of our understanding of the characters of
the indexicals and their formal interrelations; we don't always know the
latter because we don't always know the necessary extralinguistic facts,
for example, who is speaking, when, and where. The contrast between the
two kinds of knowledge lends support to the view that what we know in
the former case is linguistic or semantic. Although the linguistic doctrine
of a priori knowledge may not be true in general, it tells exactly the right
story for these sentences. In sum, the rule of character (I) is, or is a sig-
nicant part of, the linguistic meaning of the indexical `I' and what the
speaker knows about `I' as part of his semantic competence.19
Like the Old theorist's denotational content C for singular terms that
provides conditions that its referent must satisfy or t, indexical charac-
ters prima facie also seem to specify semantic conditions that their direct
referents must satisfyfor example, being the speaker or agent of the
context of the utterance of `I'. The only dierence between them appears
to be where their respective conditions are laid down: at the level of
character or at that of content. So, if satisfaction is the crux of the
denotational mechanism of indirect reference determination, an indexical
character would seem not to be very dierent from the content of an
eternal denotational term. The mechanism for xing the indexical referent
looks like satisfaction of a descriptive condition (e.g., being the speaker or
agent) but (only) in its context of utterance. Put dierently, indexicals are
directly referential, in that with respect to their evaluation at a circum-
stance, their content is not indirectly referential. But they are not thor-
oughly nondenotational, in that a mechanism like denotation appears to
be operative at the level of their character applied to their context.
Despite this appearance, I want to argue that indexical characters do
not determine their direct referents, or contents, through denotation. The
reason, in brief, is that the condition explicitly spelled out in the rule of
character for an indexical, such as `the agent (time, location, etc.) of the
context', fails to individuate a unique referent in the universe for each
utterance of the indexical (e.g., what is the context, referred to in the rule?)
and no strategy that appeals to the context to pick up the slack succeeds
in a way compatible with the requirements of denotation.20 To avoid a
90 Chapter 3

potential misunderstanding, note that this is not to say that an indexical

determines its referent denotationally only if it determines it fully and
exclusively denotationally, in terms of satisfaction of explicitly formu-
lated, purely qualitative individuating conditions. Such a condition would
ignore the obvious fact that indexicals are context-dependent; hence, even
if indexicals are denotational, context must play some role. The objection
instead is that although the rule of character requires contextual sup-
plementation to be denotational, the way in which the context can sup-
plement a rule like (I) is not by furnishing the materials for denotation,
that is, satisfaction of linguistic descriptive conditions. Let me explain.
All the plausible strategies to supplement (I) appeal to the extra-
linguistic context either to ``complete'' the incomplete description or to
supply an appropriate restriction on the universe of quantication. One
way the context can do these jobs is by adding more specic descriptive
content to the incomplete description or by appending an antecedent
condition to the quantiers. But if there are any descriptively adequate
supplementations or restrictions in a given context, there will also be more
than oneamong which there may be no saying which the given speaker
had in mind. Potential supplementations and restrictions also presumably
vary with the speaker. Both possibilities undermine the status of rules like
(I) to be a kind of linguistic meaning.
A second way context can enter is by ``supplying'' an individual or
thing that ``anchors'' the description, for example, by converting the
original to something like `the agent of the context of utterance of that
token of ``I'' ' whose content would contain the actual token of `I', a spa-
tiotemporal sound event. But this alternative runs contrary to the general
idea of denotation according to which the condition to be satised be
conceptualized or qualitative. And even if we overlook that issue, the so-
lution at best substitutes one problem for another. We want to explain
how context enables us to determine a referent for the incomplete de-
scription `the agent (time) of the context'. Now we are told that the con-
text ``supplies'' an anchoring individual for the demonstrative phrase `that
token of ``I''.' But if context can ``supply'' a referent for this demonstra-
tive phrase, why can't it directly ``supply'' an individual for the indexical
`I' without invoking a description, period?
Indeed maybe it does. That is, perhaps the context just ``supplies'' an
individual for the indexical without invoking the satisfaction of any de-
scription. To spell this out, we must say a bit more about the notion of a
context or, more accurately, about two conceptions of context current in
the literature. These run in parallel with Kaplan's distinction, mentioned
Themes from Demonstratives 91

in chapter 1, Section II, between utterances, or full-bodied speech-acts,

and occurrences of sentence(-type)s-in-contexts.
The rst notion of context is the idea of the ``total speech situation''
that includes ``everything'' relevant to the ``acceptable interpretation'' of a
particular utterance. Since dierent utterances call upon dierent kinds
of features for their respective interpretations, we must look rst at each
utterance to construct its context of this kind. And because we are dealing
with actual utterances, real-time speech acts, the context must take into
account a wide variety of factors that determine their ``success,'' only
some of which are semantic while others are psychological and physical.
The second idea of context, Kaplan's, is an abstraction from this idea
of the total speech situation. It accounts only for semantic context-
dependence from among the multiple kinds that the more heterogeneous
idea of a speech situation aims to explain. On the other hand, this notion
of context is concerned not simply with the assignment of truth-conditions
to individual sentences, but with forms of inference and with the validity
of (sets of ) linguistic types.21 And to account for the simultaneous
co-occurrence of multiple propositions in one context, this semantics-
oriented notion of context must take liberties that the idea of context
appropriate to full-blown utterances does not. What, then, belong to these
semantic contexts, and what constraints govern them? To answer this
question, we should remind ourselves of why semantics needs contexts
contexts in addition to the circumstances necessary for the evaluation of
all sentences.
The justication for contexts is that certain sentence-types have se-
mantic properties that reect their specic structure rather than the more
general structure of circumstances. For example, each occurrence of `I am
actually here now' expresses a contingent propositionevaluated at some
possible circumstances it is true, at others falsebut each occurrence of
the sentence is true in the circumstance of its context. Moreover, each
occurrence is true in its context in virtue of the fact that, in any context,
the individual directly referred to by the indexical `I' (whoever she may
be) is in the position (wherever that may be) directly referred to by the
indexical `here' at the time (whenever that may be) directly referred
to by the indexical `now' in the circumstance (whatever that may be)
directly referred to by the indexical `actually'. So, the semantics singles
out contexts as a class of structures because they explain the special
semantic properties of certain sentences. But if that is their justication,
they should consist of exactly those elements that account for those
92 Chapter 3

semantic properties that are not captured by the more general idea of
circumstances, relative to which we evaluate the contents of occurrences
of sentence(-token)s. These elements are (i) values for parameters corre-
sponding to the indexicals in the language, which (ii) are interrelated
according to certain structural constraints. For example, in every context
the value of the `I' parameter must be (located) at the value of the `Here'
parameter at the value of the `Now' parameter in the value of the `Actual
(circumstance)' parameter.22 In sum, a (minimal) context is any sequence
of four elements, an individual i, a time t, a possible world w, and a po-
sition p, such that (a) i is located at p at hw, ti and (b) i exists in hw, ti.23
To return now to how the character of an indexical determines its ref-
erence: Suppose that a context just is (or supplies) such a sequence of
values for parameters corresponding to the indexicals. When an indexical
occurs in a context, the present proposal is that we do not survey the
context, on the model of denotation, to see which of its elements satises
the condition stated in its rule of character. Instead, in situating the oc-
currence of the indexical in the contextby virtue of the fact that the
indexical occurs in the contextit is ipso facto assigned the value of its
parameter in that context, just as values are assigned to variables. The
only dierence between indexicals and variables is that the assignment of
a value to a variable is entirely unconstrained by the individual's semantic
knowledgehence, it is presemanticwhereas the value-assignment to
the indexical in each context is semantically constrained: for `I' to an
agent-value, for `now' to a time-value, and so on. But denotation, or sat-
isfaction of conditions, plays no part among these constraints. Indexicals,
then, are thoroughly nondenotational. Howeverin contrast to variables,
constants, and proper names whose assignments are semantically un-
constrainedindexicals are parametric, ``lled'' by the values of their
semantically determined contextual parameters.
To summarize: (1) An indexical is a directly referential term. Each
token of an indexical type is a rigid designator of the same referent at all
circumstances, and its content is that individual referent rather than a
representation whose conceptual conditions are evaluated at circum-
stances to determine that individual as a denotation. (2) Its character is its
meaning, an object of the speaker's semantic competence. (3) The char-
acter also does not determine the direct referent, or its content (in its
context), by way of denotation, via the satisfaction of conditions. Instead
(4) the indexical is a parametric expression ``lled'' by the value of its
corresponding contextual parameter. This assignment is semantically con-
strained, but the indexical, like a variable, is thoroughly nondenotational.
Themes from Demonstratives 93

V Dthat-descriptions and Complete Demonstratives

I now turn to the dthat-description, the second main theme from the
theory of demonstratives that shapes my semantic account of metaphor.
Dthat-descriptions, like indexicals, are directly referential terms. (1) They
are rigid designators and (2) their content is their individual referent; that
is, their content is not a conceptualized representation that, evaluated at a
circumstance, determines the referent by denotation. However, the way in
which the character of a dthat-description determines its direct referent
which, as with indexicals, is its contentis in sharp contrast to the para-
metric character of an indexical. As we shall see, denotation does the
main work for the character of the dthat-description; therefore, dthat-
descriptions are not thoroughly nondenotational.
Looking ahead to metaphor, I emphasize this dierence because the
dthat-description provides us with the model for our treatment of the se-
mantic context-dependency of a metaphor. However, the way the char-
acter of a metaphor expresses, or determines, its content, or property, in
a context is, unlike dthat-descriptions, not by way of denotation. The
characters of metaphors, like those of proper indexicals, are parametric.
For this reason, metaphors are a hybrid of the two.
In his rst paper on demonstratives, ``Dthat,'' Kaplan was inspired to
create the dthat-description through reection on Donnellan's referential
use of the denite description. However, like many other things we
humans parent, our creations often take on a life of their own, exceeding
our own greatest expectations about their futures. By the time he wrote
``Demonstratives'' in the '70s, Kaplan writes that he ``regards [the] `dthat'
operator as representing the general case of a demonstrative'' (Kaplan
1989a, 527), and in the formal ``Logic of Demonstratives'' the demon-
strations that complete pure demonstratives are indeed treated exactly as
if they were linguistic singular terms, which, in turn, bears out Kaplan's
quoted claim. Nonetheless, despite these considerable parallels, my own
view is that there are great disanalogies between complete demonstratives
(with their extralinguistic demonstrations) and dthat-descriptions, dis-
analogies that legislate against taking the `dthat' operator as representing
the general case of a demonstrative. Here is not the place to pursue that
discussion.24 However, in order to explain Kaplan's insight that moti-
vated the dthat-description, it will be valuable to place it in the context of
his analysis of complete demonstratives and, in particular, his Fregean
theory of demonstrations. I'll begin, then, with the common ground
shared by complete demonstratives and dthat-descriptions, bracketing
94 Chapter 3

their many dierences and many of the complications that arise in the
explanation of complete demonstratives per se.
As I said at the beginning of this chapter, a demonstrative must be
completed by a demonstration in order to determine a referent; without
the demonstration, the pure demonstrative is incomplete. Let us call the
individual demonstrated by the accompanying demonstration its demon-
stratum. The rule of character for a complete demonstrative `That[D]',
analogous to (I) for `I', is, then, a rule like (D):
(D) In any context c, each utterance of `That[D]' directly refers to the
demonstratum of the completing demonstration D in c, and otherwise
directly refers to nothing.
(D) raises two main questions. First, what is a demonstration? Second,
how should we understand the claim that the demonstration D completes
the pure demonstrative `That'? What is the syntax and semantics of
`That[D]'? Is it syntactically and/or semantically simple or complex? I'll
return to this second question at the end of section VIII in connection
with dthat-descriptions.25 The answer to the rst question, which is the
immediate background to the idea of a dthat-description, is complicated
by the fact that two dierent notions of a demonstration surface in cur-
rent accounts of demonstratives. Kaplan mentions both notions but
unfortunately he equivocates between them.
The rst idea of a demonstration is the ordinary idea of a bodily ges-
ture, action, or cue, the paradigm of which is the act of pointing at an
object. This is also the popular conception of a demonstration; witness the
Random House Dictionary entry for `That': ``the person or thing pointed
out or mentioned.'' This act is accompanied by, in Kaplan's words, a
``directing intention,'' and it is a further question whether what really
determines the demonstratum is the bodily act itself or the intentional act
or state that the bodily act, for communicative purposes, merely exter-
nalizes. I also include under this rst idea extralinguistic events, happen-
ings, or other contextual occurrences that make something the focus of
attention in place of a gesture of pointing, for example, the fact that the
thing is in the spotlight or on a podium. The second notion of a demon-
stration, which may have been rst proposed by Husserl, is introduced
by Kaplan when he describes a demonstration as ``typically, though not
invariably, a (visual) presentation of a local object discriminated by a
pointing.''26 Here I take the demonstration to be the ``presentation''
whatever that isalthough the object presented is also ``discriminated''
by a pointing.
Themes from Demonstratives 95

Prima facie these two notions of a demonstration are mutually com-

patible. Indeed the typical complete perceptual demonstrative involves
both: a gesture (like pointing) to an object perceptually presented in the
context in a certain way. However, a number of recent philosophers have
tried to pull them apart, seizing on one to the exclusion of the other as the
``real'' demonstration, depending on the problem the particular author
thinks is primary. In general the gesture is given the role to x the refer-
ent, while the presentation accounts for cognitive signicance. Of course,
if both problems must be solved and each kind of demonstration accounts
at most for one, any adequate theory of demonstrations must accommo-
date both. This is indeed closest to my own view, although I shall not try
to defend or elaborate it here. Instead I'll concentrate on demonstrations
as presentations, which will lead us to dthat-descriptions.

VI Demonstrations as Presentations

The motivation for the idea of a demonstration as a presentation is the

same set of considerations, or data, that prompted the fourth role (and, to
some degree, the fth) that the Old theory assigned to its associated con-
ceptual representations: to explain cognitive phenomena like informa-
tiveness (as in Frege's identity puzzle) by positing modes of presentation,
or ways of thinking, of their referents. Kaplan, parodying Frege, gives an
example of an analogous puzzle for demonstratives. How should we ex-
plain the cognitive dierence between the informative utterance of
(2) That [the speaker points at Venus in the morning] That [the
speaker points at Venus in the evening],
as opposed to the uninformative utterance of
(3) That [the speaker points at Venus in the morning] That [the
speaker points at Venus in the morning].
Since all the complete demonstratives in both (2) and (3) directly co-refer,
they not only have the same referent but also the same content; hence the
dierence in their information-values must be sought elsewhere.
Back in section I, I said that Kaplan shares the Old theory's per-
spectivalist epistemology: that we always refer from a perspective or think
of things in a particular way. One way to capture these perspectives or
ways of thinking of things is to build them into or encode them in asso-
ciated representations, be it at the level of content, truth-conditions, sense,
or (as we'll see in a moment) character. This strategy I shall call repre-
96 Chapter 3

sentationalist. However, a second, and simpler, way to account for certain

dierences of perspective on a co-referent is by taking the referent to be
thought of (or by taking tokens of the co-referring expression to occur) in
dierent extralinguistic contexts. Thus an utterance may be informative
because it expresses the discovery that the referent of a term occurring in
one context is identical to the referent of the same term (type) occurring in
another context. Dierences of informativeness would correspond, then,
to dierences between extralinguistic contexts. I'll call this second strategy
of capturing a speaker's cognitive perspective contextualist.
The most plausible, and simplest, explanation of the cognitive dier-
ence between (2) and (3) is contextualist: In (2) one demonstrative is
uttered in the morning in one season, the other in the evening in another
season. In (3) they are both uttered at the same place and time. What
makes (2) informative is that we thereby learn that the thing referred to
at one time and place with a certain appearance is identical to the thing
referred to at a dierent time and place with a dierent appearance.
Likewise, if there were analogous puzzles exclusively involving proper
indexicals such as the uninformative
(4) I I
versus the informative
(5) I you
or the uninformative
(6) Tomorrow tomorrow
versus the informative
(7) Tomorrow today,
the most plausible explanation for the informative utterances is con-
textualist. Since it is reasonable to assume that in any one (normal) con-
text, no single thing can ll two parameters, say, simultaneously be agent
and addressee or both the day of utterance and the next or previous one,
the anking co-referential indexicals must occur in dierent contexts in
order for their utterances ((5) and (7)) to be true. To be sure, we generally
require a sentence-occurrence to be evaluated in one and only one con-
text, but for these to be true, the context must shift mid-sentence. And if
their truth requires these sentence-occurrences to split themselves between
two contexts, we can also, and most easily, explain their cognitive signif-
icance in terms of cross-contextual continuities they track.27 It would be
gratuitous to incorporate dierent perspectival representations in their
structure only in order to account for their informativeness.28
Themes from Demonstratives 97

However, it is also evidentand this was surely Kaplan's idea

that there are informative identities involving complete demonstratives
that occur within a single context and that cannot be explained cross-
contextually. Contrast, for example, the informative identity
(8) He [points to a newspaper photograph of Superman, midair dressed
in his blue tights] he [points to Clark Kent, dressed in mild-mannered
suit, standing before us]
with the uninformative
(9) He [points to a newspaper photograph of Superman, midair dressed
in his blue tights] he [points to the same newspaper photograph of
Superman, midair dressed in his blue tights].
Here the informativeness of (8), as opposed to (9), that goes beyond
their common content must be explained relative to a single context, in
which case the natural explanation would be to build representational
dierences into the complete demonstratives.29 But how? (8) seems to
be informative because of the dierent visual features, or appearances,
``associated with'' the co-demonstratum of the two occurrences of `he'.
But what is the nature of this association, and is it a representational,
semantically individuated dierence between the complete demonstra-
tives? The two occurrences of the pure demonstrative (`he') look like
tokens of the same type, with the same character, hence the same lin-
guistically determined mode of presentation. The gesture, the act of
pointing at the object, is also the same. All that changes is the visual ap-
pearance of the object pointed at. But that is a dierence prima facie in
what is being represented, not in the representing entity.
Kaplan's proposal is to build these dierences in the appearances of the
co-demonstratum into the representational structures of the complete
demonstratives. The dierences in appearance are not just dierent modes
of (perceptual) presentation; they constitute dierent presentations, hence
dierent demonstrations. But what is a presentation? Three basic ideas
underlie Kaplan's notion.
1. Let a presentation be enough like a linguistic representation (read: re-
presentation) to be analyzed with a sense and referent or, in Kaplan's own
``corrected'' version, with a character, content, and referent. The thing of
which it is a presentation can be treated as its referent and the mode by
which it presents that individual can be treated as its sense, or character/
content. (These last two will be distinguished where the presentation
``contains'' context-sensitive elements like indexicals.) Kaplan calls this
the ``Fregean Theory of Demonstrations.''30
98 Chapter 3

2. Presentations, like linguistic expressions, admit a type-token distinc-

tion (525). Hence presentations (types) admit replication.31
3. Presentations are not rigid designators. A typical presentation is of a
particular thing, but it is not necessary that it be of that particular thing
(or of anything, to allow for the possibility of vacuous presentations, say,
in cases of hallucination). Suppose I am presented with my Macintosh
(serial) a100C54X129. That machine has a certain appearance, indeed
the same appearance as myriads of other Macs. And in some other cir-
cumstance another Mac (say, serial a100C54X130) might be the only
thing that matches that presentation. In that counterfactual circum-
stance, the viewer would stand in the very same perceptual relation to
a100C54X130 that in fact I stand to a100C54X129. Which computer is
presented depends, not only on features of the presentation, but also on
the contingent circumstances of the presentation. In dierent circum-
stances the same presentation might present dierent, though perceptually
indiscriminable, things.
To capture these ideas, Kaplan models presentations after a partic-
ular type of linguistic representation, the nonrigid denite description.
Thus (9)
(9) He [P the midget with a mustacheP ] is a terrorist32
contains the pure demonstrative `he' completed by a presentation in
which, say, Y is presented with the same qualitative visual properties
which he is described as possessing in (10)
(10) The midget with a mustache is a terrorist
that is, as if the presentation were itself a linguistic expression. The one
semantically signicant dierence between (9) and (10) is that in (10) the
nonrigid descriptive condition is part of the content of the utterance; in (9)
the content of the complete demonstrative is the individual Y. The quali-
tative properties of the presentation belong to the character that deter-
mines the content without being part of it. So, the proposition expressed
by (10) is
(10 0 ) hThe M, Ti
where the italicized letters M and T stand for the properties of being a
mustached midget and being a terrorist, whereas the proposition
expressed by (9) is
(9 0 ) hY, Ti
Themes from Demonstratives 99

or
(9 00 ) h{He[P The MP ]}(c), Ti
in which the component corresponding to the complete demonstrative is
Y rather than the qualitative conditions that constitute its presentation.
It is the nonrigid content of the denite description that enables the
Fregean to account for its cognitive signicance and solve his puzzles.33
By giving presentations the same semantic structure, Kaplan wants to
solve the analogous problems for the directly referential terms for which
Frege's own solution will not work. Slightly revising Kaplan (see note 31),
let's take a presentation to have the following ``standard form'':
(P) The individual that has appearance A
where an appearance, Kaplan continues, is
something like a picture with a little arrow pointing to the relevant subject. Trying
to put it into words, a particular demonstration might come out like: `The
brightest visible heavenly body'. (Kaplan 1989a, 526)
Let's suppose that a linguistic description can serve as a surrogate for the
appearance.34 The character of a complete demonstrative such as (11)
(11) That[P The brightest visible heavenly bodyP ]
is composed from the characters of its parts (or from those of its parts
that have their own characters, like the embedded presentation). So, the
character of this complete demonstrative (type) is the rule that each of
its tokens directly refers to the thing (in its context) that is the brightest
visible heavenly body. Presentations with dierent visual properties sub-
stituted for the appearance, A, have dierent characters and, by compo-
sition, so will the complete demonstratives built from them. The only
dierence between the character of the complete demonstrative and of its
constituent presentation is that the former is directly referential and the
latter is not even rigid. Therefore, there can be two presentations with
dierent contents in the same context, but, if they are presentations of the
same thing, their respective complete demonstratives will have the same
content, that is, a co-referent. This excludes an explanation of the dier-
ence in the informativeness of the two complete demonstratives at the
level of their respective contents. Instead, Kaplan argues, we can explain
it by their respective characters. For two presentations have dierent
contents in the same context i they also have dierent characters, and
two presentations have dierent characters i their two respective com-
plete demonstratives also have dierent characters. Hence dierent visual
100 Chapter 3

contents of two presentations will ultimately surface in dierent charac-

ters for their respective complete demonstratives. qed, Frege's puzzle for
demonstratives is solved!35

VII Dthat-descriptions

The Fregean Theory of Demonstrations treats the visual presentation D

as if it were a denite description, assigning it a character, content, and
referent.36 Having drawn this parallel, Kaplan next extends it in the
opposite direction. Just as D completes the pure demonstrative `That',
Kaplan introduces a special demonstrative symbol `Dthat', which is
``completed'' by a (nonrigid) denite description F.37 `Dthat' has two
eects. First, it takes a singular term F, whose content is a complex of
descriptive conditions, and yields the directly referential dthat-description
Dthat[F] whose content is its individual referent. Second, it takes a sin-
gular term F whose character (assuming it contains no indexicals) is
constant and yields the dthat-description whose character is nonconstant
determining dierent contents ( direct referents) in dierent contexts.
Thus the description `the midget with a mustache' contributes the same
complex of properties as its content in all contexts, but `Dthat[`The
midget with a mustache']' contributes dierent contents, that is, direct
referents, in dierent contexts, depending on who in fact is the midget
with a mustache in the circumstances of its context. The total eect of
`Dthat[`The midget with a mustache']' is to make the content of the
embedded description part, not of what is said by its utterances, but of the
character of its utterance, which determines, relative to a context, what it
says. Thus the character rule for dthat-descriptions:
(Dthat) For every context c and for every denite description F, an
occurrence of `Dthat[F]' in c directly refers to the unique individual (if
there is one) denoted by F in the circumstance of c, and to no one
otherwise.
Although both eects of `Dthat' will be important for the application to
metaphor, it is the secondthe fact that `Dthat' generates terms with
nonconstant characters from terms with constant charactersthat I shall
work out in greatest detail in chapter 4.
I should emphasize one feature of dthat-descriptions in anticipation of
my account of metaphor. When a speaker learns the rule (Dthat), what he
learns is not the character of a single expression (type) in the language
as he does when he learns, say, (I) for `I'. What he learns is a schematic
Themes from Demonstratives 101

rule whose instances are the result of substituting particular denite

descriptions in the language for the metalinguistic variable F. That is, he
learns a rule that enables him to interpret any well-formed denite de-
scription demonstratively; he acquires a skill of interpretation rather than
an individual piece of vocabulary. Although we'll see in a moment some
specic diculties that attend this idea, this aspect of `Dthat' is also what
lies behind the claim that it is an operator on the description rather than a
well-formed term in its own right.
The one respect in which dthat-descriptions turn out not to be a model
for metaphors, as I mentioned earlier, is that their characters, as the rule
(Dthat) shows, employ denotation: At the content generation-stage, it is
denotation, satisfaction of the descriptive conditions given in the charac-
ter of the embedded description F, that does the work in xing the direct
referent of `Dthat[F]'. Thus dthat-descriptions, although directly referen-
tial, are not thoroughly nondenotational. Metaphors, I'll argue in chapter
4, are in this respect like indexicals, which, being parametric, are not only
directly referential but also thoroughly nondenotational.
Parallel to the thoroughlynot thoroughly nondenotational distinction,
there is a second dierence between indexicals and dthat-descriptions that
also bears on metaphorindeed on dierences between metaphors. The
characters of indexicals are de jureby their very naturenonconstant,
whereas the characters of dthat-descriptions are only de facto noncon-
stant. Take indexicals rst. The value of an indexical parameter in any
one context is entirely independent of its value in any other; for example,
the fact that I, JS, am the value of the speaker parameter for `I' in one
context is entirely independent of the fact that DD is the value in another.
Of course, the same thing could turn out, as a matter of pure chance, to
be the value of the parameter in more than one context, but the characters
of the indexicals legislate, as it were, that we look only at the values of the
parameter in each context one by one. If a value is preserved across con-
texts, it is nothing more than accidental homonymy. It follows that, if
indexicals are parametric, the character of an indexical must be non-
constant, sensitive to each context, one by one. The character of a dthat-
description, on the other hand, is (only) de facto nonconstant, because it
is only as a matter of fact that its constituent description does not denote
the same thing in all circumstances. If the constituent description F were
such that it turned out (as a matter of fact, physical or metaphysical) to
denote the same thing in all circumstances, then it would also ipso facto
denote the same thing in (the circumstances of ) all contexts. In which
case, the dthat-description `Dthat[F]' would also turn out to have a de
102 Chapter 3

facto constant character. Unlike indexicals, there is nothing about the very
nature of dthat-descriptions that renders their characters nonconstant.
I emphasize this last point because the de facto context-dependence of a
dthat-description is compatible with its being suciently rich descriptively
or conceptually to uniquely determine its referent at a circumstance
without needing extralinguistic or nonconceptualized contextual aids.
Perhaps it would be more accurate to speak of its context-sensitivity
rather than context-dependence, but the role of the context in the seman-
tics of the dthat-description is limited to the fact that it is the circumstance
of its context, namely, the actual circumstance, that is the relevant cir-
cumstance for determining (by denotation) its directly referential content.
I'll argue in chapter 8 that the nonconstant character of some meta-
phors is more like the de jure nonconstant character of indexicals whereas
that of other metaphors is more like the de facto nonconstant character of
dthat-descriptions. In particular, I'll argue that one kind of dead meta-
phor is a metaphor whose character is nonconstant and context-dependent
but only de facto; the liveliest metaphors are those whose characters are
de jure nonconstant.
Finally, I want to return to one issue that arises in part because of the
not thoroughly nondenotational character of dthat-descriptions, an issue
that, again, marks a dierence (as we'll see in ch. 6) between dthat-
descriptions and metaphors. This is the question we broached earlier in
connection with the character rule (D) for complete demonstratives: If
the demonstration D ``completes'' the pure demonstrative `That,' is the
resulting expression `That[D]' syntactically and/or semantically simple or
complex? In the case of the dthat-description, where the issues are sharper
(since the constituents are all linguistic), we can appreciate the question in
all its complexity. There are strong reasons for holding both that it must
be syntactically simple or unstructured and that it must be syntactically
complex or articulated.38
On the one hand, a dthat-description is a directly referential term
whose content is simple, namely, its individual referent, as opposed to a
merely rigid designator whose content is a complex conceptualized repre-
sentation that rigidly designates the same individual at all circumstances.
Therefore, mirroring its semantic simplicity, its syntax ought to be simple
too. For this reason, Kaplan sometimes writes that despite the fact that
`dthat' is provably equivalent in his formalized logic of demonstratives to
a rigidifying operator syntactically completed by the description as an
operand (1989a, 552, Remark 13), it should be analyzed as a ``syntacti-
cally complete term that requires no syntactical completion by an oper-
and'' (Kaplan 1989b, 581).39
Themes from Demonstratives 103

On the other hand, there are two good reasons to articulate the char-
acter of the constituent description as part of the character of the com-
plete dthat-description. First, as we saw (in section VI), the motivation
for positing a constituent presentation in the complete demonstrative was
to enable us to account for its cognitive signicance, the kind of infor-
mation that solves puzzles like Frege's. Similarly, the character of the
embedded term in the dthat-description carries the mode of presentation
that enables us to distinguish informative from uninformative utterances
of, say, identity statements anked by dthat-descriptions, despite the fact
that they have the same content. However, to recover this character-istic
information carried by dthat-descriptions, they must have constituent
syntactic structure. Second, the articulated character of the embedded
description determines how the direct referent of the complete dthat-
description is xed, namely, by denotation. But if the description F must
be discernible as an autonomous denoting unit within `Dthat[F]', there
must also be syntactic structure to the whole composite or complex
expression.
These two reasons suggest that the characters of dthat-descriptions are
articulated and, if their characters are structured, their syntax should be
too.40 Hence the dthat-description is pulled in two directions: Its two
motivating ideas lead to the contradictory conclusion that it ought to be
both semantically and syntactically simple and complex.41 One might
further conclude that, ingenious as the invention of `Dthat' is, it is an at-
tempt to do the impossible. Should we therefore abandon it?
The last verdict may be too harsh.42 There are other constructions in
natural language for which philosophers have given competing reasons to
hold that they are both syntactically and/or semantically structured and
unstructured; examples are quotation, both direct and indirect (that-
clauses). In those cases, we certainly wouldn't conclude that we should
abandon the constructions. Despite similar theoretical tensions, we
wouldn't condemn the natural language constructions as attempts to do
the impossible. Rather, the conicting considerations show that there re-
main unresolved problems in our understanding, that the fault lies with
inadequate theories rather than the phenomena.
Here, too, perhaps we should charitably conclude that `Dthat' is a
successful representation for how we interpret, or use, (nondemonstrative)
expressions demonstratively, though we do not yet fully understand how
it succeeds at this task. In this spirit we'll also see how `Dthat' provides us
with a technique of great expressive power to import context-sensitivity
into interpretation where previously there was none. `Dthat' gives us a
way to understand and explain demonstrative interpretations (or uses) of
104 Chapter 3

arbitrary expressions in addition to the explicit demonstrative and index-

ical expressions. Demonstrativity is no longer an exceptional, idiosyn-
cratic feature of a handful of isolated words, but a general mode of
interpretation that can be foundor introducedthroughout all of lan-
guage. Nor is it limited to the demonstrative mode of interpretation
narrowly character-ized by `Dthat'. In the next chapter I'll argue that
metaphorical interpretation is yet a further mode of the same general type
of context-sensitive interpretation of language.
Chapter 4
Knowledge of Metaphor

With Kaplan's semantics for demonstratives in hand, I am now in a po-

sition to make good on my promissory notes for metaphor. The ``inter-
pretations'' of a metaphor (type) are the contents its tokens express in
their respective contexts. Since there is an unlimited, or not antecedently
xed, number of dierent possible contexts in which those tokens can
occur, there is an unlimited number of dierent possible contents those
tokens can express metaphorically. And because each is a function of its
extralinguistic context of utterance, the speaker's knowledge of their con-
tents cannot be a matter simply of her linguistic competence. Yet, as I
illustrated in chapter 1, examples (1)(10), the variations in content seem
to follow a pattern of corresponding variations in their respective con-
texts; and as we saw in chapter 2, there are also constraints on the possible
metaphorical interpretations (in dierent contexts) of one expression
(type). Furthermore, those constraints look like those that govern specic
kinds of literal interpretations of language. These facts suggest that a
speaker has a more abstract kind of knowledge apart from his knowledge
of the particular content of each metaphorical token in its respective
context. This more abstract piece of knowledge is the character of the
metaphor.
As with the character of any expression, the character of an expression
interpreted metaphorically is a function, or rule, that determines, for each
context, its content in that context. Like the nonconstant characters of
demonstratives, which determine dierent contents relative to dierent
contexts, the character of a metaphor is nonconstant. And like the char-
acters of both indexicals and dthat-descriptions, a speaker's knowledge of
the metaphorical character of an expression (i.e., the character of the
expression interpreted metaphorically), though not of its content, falls
within her semantic knowledge; hence it is knowledge of a kind of (lin-
guistic) meaning. Yet, beyond these common themes, the character of a
106 Chapter 4

metaphor is a hybrid, in some of its more specic features more indexical-

like, in others, more dthat-description-like. I shall mention two of these
for now, beginning with a dthat-like characteristic.
First, recall that a speaker's knowledge of the rule of character (Dthat)
for the dthat-description is schematic: The rule applies to any description
that can be demonstratively interpreted. Hence knowledge of (Dthat) is
dierent from knowledge of the character, or meaning, of a single ex-
pression (type) in the language, such as the character of `I'. Similarly,
what the speaker knows when she knows the character of a particular
metaphorically interpreted expression is an instance of a character-
schematic rule, which applies not only to the particular expression-type
interpreted metaphorically on the occasion; what she knows holds for any
expression that it is possible to interpret metaphorically. Like her knowl-
edge of dthat-descriptions, what a speaker knows in knowing the meaning
of a metaphor (not: its content in a particular context) is, like a skill of
interpretation, a rule that enables her to interpret any expression meta-
phorically, given knowledge of the relevant context.
Let me make this semantic parallel lexically explicit. Just as Kaplan
invented `Dthat [F]' to lexically represent the demonstrative interpreta-
tion of an arbitrary denite description F, I shall now create an analogue
`Mthat[F]' to lexically represent the metaphorical interpretation of an
arbitrary (literal) expression F. Given any (literal) expression F, `Mthat'
yields the (to coin a term of art) ``metaphorical expression'' `Mthat[F]'
that has a nonconstant character sensitive to a specic ``metaphorically
relevant'' feature of its context. Thus both `Dthat' and `Mthat', completed
by their respective expressions, yield lexical representations of specic
kinds of context-dependent interpretations of those expressions. Just as
the speaker who knows how to generate an interpretation, or content, for
a demonstratively interpreted denite description F knows the (non-
constant) character of the dthat-description `Dthat[F]', so a speaker who
knows how to interpret an expression F metaphoricallywho knows
what it is to generate a metaphorical content for F in a contextknows
the (nonconstant) character of the metaphorical expression `Mthat[F]'.
Second, unlike dthat-descriptions but like indexicals, the metaphorical
expression `Mthat[F]' is parametric. Its content in a context is not de-
noted by F, the expression interpreted metaphorically. Instead the con-
tent is the value of a parameter in the context to which metaphors are
sensitive, just as the content of `I' in a context is the value of the respective
parameter to which that indexical is sensitive. (I shall turn next to the
identity of the parameter for metaphor.)
Knowledge of Metaphor 107

Finally, unlike both singular indexicals and dthat-descriptions whose

content (in a context) is an individual, but like the predicate demonstra-
tive `Thus' or `is that F', the content of `Mthat[F]', in the basic case where
it occurs in predicative position, is a property or set of properties.1
Therefore, the character of `Mthat[F]' is a function from the set of con-
texts (or the set of relevant contextual parameters) to a set of (sets of )
properties rather than to a set of individuals.
This sketch of the shared and distinct structures of metaphors, dthat-
descriptions, and indexicals needs considerable lling in. Two main tasks
lie ahead: (i) to describe the contextual parameter on which a metaphori-
cal interpretation depends, and (ii) to spell out the logical structure of
metaphorical character, that is, for each expression F that is interpreted
metaphorically, the character of its corresponding metaphorical expres-
sion `Mthat[F]'. In the remainder of this chapter, I shall work out the
basic theoretical notions that constitute our knowledge of metaphor, the
specically semantic competence that underlies our ability to interpret a
metaphor. In chapter 5, I'll illustrate how this theoretical apparatus
interacts with, or is applied to, a context to yield a content; this chapter
will thereby address one of the two ways by which knowledge is (con-
veyed) by metaphor. Here, I should emphasize, my point is neither to
demonstrate nor to test the adequacy of our semantics to produce actual
interpretations; it is to show how their semantic structure places con-
straints on metaphorical interpretations. In chapter 6 I return to apply
our account of knowledge of metaphor to some outstanding problems
that we have already encountered, for example, how an expression inter-
preted metaphorically nonetheless ``has'' its literal meaning; how the
metaphorical interpretation of an expression ``depends'' on its literal
meaning; why metaphorical character is a type of meaning; and the rela-
tion of metaphor to other tropes. Finally, in chapter 7, I turn to the sec-
ond way by which knowledge is (conveyed) by metaphor, namely, by way
of its character. Here I shall show, as I promised in chapter 1, how a
metaphor can convey knowledge by its metaphorical character that is
not captured in its (literally paraphraseable) propositional content in a
context.

I The Context of a Metaphor: Presuppositions

Max Black (1962) was the rst contemporary author to recognize that
what is important for the metaphorical interpretation of an expression F
is a range of beliefs ``associated with'' F and not, or not only, its ``stan-
108 Chapter 4

dard dictionary meaning'' or ``literal sense.'' After making this observa-

tion, Black concentrated his discussion on beliefs he called ``the system of
associated commonplaces'' for F. He described these beliefs in a number
of ways: as beliefs that would be ``readily and freely'' evoked by F for the
ordinary (nonexpert) member of the linguistic community (or, in his
terms, ``culture''); as ``current platitudes'' about Fs that are ``the common
possession of the members of [the] speech community''; and as the kinds
of beliefs to which a speaker's literal use of a word ``normally commits''
him. For example, the system of associated commonplaces for `is a wolf ',
in the metaphor `man is a wolf ', is the ``system of ideas'' of being ``erce,
carnivorous, treacherous, and so on.'' This system, Black observed, is
``not sharply delineated'' (note the ``and so on'') though it is ``suciently
denite to admit of detailed enumeration'' (ibid., 4041). None of these
properties, he continued, would be properly speaking considered part of
the dictionary meaning or denition of `wolf '. But we would surely expect
special pleading by someone who instead interpreted the metaphor to ex-
press the proposition that, say, man is a vegetarian. This interpretation
would normally be evidence that the speaker is not a member of our
speech community or that she falls short of our expectations of common
knowledge among competent members of the community.2
Much of what Black says is right on mark. His denial that the meta-
phorically relevant ``knowledge'' is knowledge of the meaning or deni-
tion of F highlights the fact that the metaphorically relevant beliefs need
not be true (e.g., when the system of commonplaces for `whale' includes
the belief that a whale is a sh), not to say necessary or analytic to F. It is
much more important that the beliefs be shared and be publicly accessi-
ble, although, contrary to his phrase ``readily evoked,'' they need not be
eortlessly or instantaneously recoverable. He does not explain why he
says it, but Black is also right that the relevant beliefs must be associated
with the word F and not just with its extension or intension. Finally, he
emphasizes the systematicity of the metaphorically relevant beliefs, an
aspect of metaphorical interpretation that has drawn much recent atten-
tion, rst through the work of Goodman and lately by Lako and his
school. We shall return to this last theme in chapters 5 and 7.
Black's emphasis on ``associated commonplaces'' is, however, unfortu-
nate. Although he acknowledges in the latter part of his (1962) essay that
the description ``associated commonplaces'' only ts the ``commonest
cases'' and that ``in a poem, or a piece of sustained prose, the writer can
establish a novel pattern of implications for the literal uses of the key
expressions, prior to using them as vehicles for his metaphors'' (ibid., 43),
Knowledge of Metaphor 109

his emphasis on beliefs linked to membership in the linguistic or speech

community tends to blur the distinction he wishes to draw between mean-
ing and belief. It also suggests that, as a rule, there is one kind of pool or
system of beliefs``commonplaces''for all typical, nonpoetic meta-
phorical interpretations. This is not true. The range of potentially relevant
properties believed to be ``associated'' with the expression to be interpreted
metaphorically is highly variegated. Not all associations are necessarily
shared by all members of the community, nor are they always ``believed''
to be true. Indeed they defy any simple uniform characterization.
Some metaphorically relevant properties do belong to the denition of
F and are true of Fs, as in:
(1) Shamir was a midget among Israeli leaders
that is, was very small (relative to humans), which is true and even de-
nitional for `midget'. But other properties may be neither true of nor even
believed to be true of F's extension; indeed they may be known to be false
of F's extension. For example,
(2) Kripke is an alchemist of ideas
can be interpreted metaphorically to say that Kripke can produce ex-
tremely valuable ideas from the most common, obvious, and prima facie
trivial intuitions despite the fact that we know that no alchemist can really
produce gold from base metals. In some of these cases that rest on false-
hoods, the feature is stereotypical and belongs to a socially shared
``meaning'' for the term; for example,
(3) Saddam is a gorilla
says that he is erce and violent, even though we know that in fact gorillas
are shy, timid, and sensitive (and violent only when violently provoked).3
In other cases of socially shared ``meaning,'' the metaphor is based on
what we might call, borrowing a term from Marc Crimmins, ``normal'' as
opposed to ``idiosyncratic'' features.4 A normal feature of an object re-
ferred to by an expression F is a feature that belongs to (more or less)
community-wide shared knowledge about Fs given a common way of life,
for example, a feature associated with F that is related to the function of
F within the general human cognitive system. Thus temperature terms
like `hot' and `cold' have normal features correlated with our standard
sensory modes for identifying temperatures, and these features are rou-
tinely exploited when the terms are metaphorically interpreted to apply
to, say, emotional states or music. As Crimmins speculates, it is more
plausible to expect expressions for properties and relations to have normal
110 Chapter 4

features associated with them than names of individualsalthough

names for ``public gures,'' past or present, universal or local, are also
likely to have normal features, for example,
(4) Clinton is a Kennedy.
On the other hand, a feature is idiosyncratic when it is associated with
an expression for only an isolated individual, sometimes only through
psychological chains of which she may not even be consciously aware.
For example, the psychoanalyst Benjamin B. Rubinstein describes the
case of a young woman whose father had administered enemas to her
until adolescence and, as an adult, could not achieve an orgasm except
when masturbating. Her most frequent masturbation fantasy was that she
was being given an enema by a young doctor in a hospital. Once, upon
being told by her analyst that she apparently preferred an enema to sexual
intercourse, ``she seemed surprised, as if she had never heard this before''
(despite having had the connection pointed out to her on numerous pre-
vious occasions), and exclaimed:
(5) Oh, that's awfulI have to wash that out, that desire.
According to Rubinstein, ``the remarkable thing is that the patient, who
was intelligent and alert, was completely unaware of the metaphorical
character of [her] utterance,'' notwithstanding the fact ``that, as a rule, her
speech was straighti.e., nonmetaphorical.''5
Now, normal (as well as many stereotypical) features are learned, or
acquired, by individuals in processes not easily distinguished from those
in which they learn the words associated with them. They are shared
within a community, and knowledge of certain normal features may even
be a qualication to be a competent member of the community. There-
fore, one might be tempted to consider a normal feature of its referent as
part of the linguistic meaning of a term. On the other hand, features are
(normally) normal relative only to a given linguistic community, varying
from one to another. Therefore, one ought to resist, I think, the tempta-
tion to think of them as truly linguisticthat is, linguistic in the same
sense in which knowledge of character proper is linguistic. We should
distinguish the kind of knowledge that is the object of a speaker's lan-
guage facultyhis knowledge of language proper, knowledge of meaning
that interacts with the other components of the language faculty, syntax
and phonologyfrom the knowledge he possesses in virtue of belonging
to a linguistic community, the knowledge we reasonably expect every
speaker in the (linguistic) community to possess on pain of being deemed
``incompetent'' in a broader sense. Knowledge of normal features asso-
Knowledge of Metaphor 111

ciated with expressions (where there are such notions) counts as the latter
but not the former. Although we may reasonably expect every speaker of
English to know the normal features associated with a term, if he doesn't,
he is not necessarily ignorant of something in the syntax, semantics, pho-
nology, or lexical meaning of the word. Consequently I will consider even
normal features as extralinguistic knowledge.
The features I have discussed thus far are relatively stable across utter-
ances, given the context of a xed linguistic community. But many meta-
phorically relevant properties attach to an expression only in a context,
lasting only for the local duration. In the next chapter, I shall discuss at
length one class of metaphors I call exemplication metaphors that are
highly context sensitive in this way. But they are not unique. Poets fre-
quently build up associations around a word in order to exploit them in
later metaphorical uses of the same word within the limited scope of one
poem.6 For example, in Antony and Cleopatra, Shakespeare rst refers to
Antony using `the sun' in order to contrast him with Cleopatra who is
described as `the moon'. Like the sun, Antony has a xed, predictable,
reliable, steady course; Cleopatra, like the moon, is erratic and unreliable.
And just as the moon derives its light from the sun, so Cleopatra is sub-
ordinate to and dependent on Antony. Later in the play, having already
singled out this association of the notion of a regular, reliable, xed
course with `the sun', Shakespeare then describes the death of Antony in
terms of the violent, cosmic catastrophe that would result if the sun were
to wander out of its xed orbit. Thus the two guards:
(6) 2. Guard: The Starre is falne.
1. Guard: And Time is at his Period. [Antony and Cleopatra, IV,
xiv, 129f.]
Here the `falling'the disappearance or `falling out of course' of the
sunmeans the end (or `period') of Time, the principle of organization
governed by the regular motion of the sun.7 So, having made one, not
obvious (and, by our current lights, false) association with `the sun',
Shakespeare turns it around in the later use. The same idea is emphasized
by the literary critic, Mario L. D'Avanzo, while commenting on Keats:
Through metaphor words take on a world of possible connotations and, depend-
ing upon the precision with which they are ordered in the poem, may precipitate
new and surprising networks of reference.8

In yet other examples, the history of a metaphor's interpretations may

also become one of the properties associated with the expression as it is
interpreted metaphorically on a later occasion. The Marxist political phi-
112 Chapter 4

losopher G. A. Cohen reports how another philosopher friend ``roused

[him] from what had been [his] dogmatic socialist slumber'' by ``hitting''
him with an argument that was to appear in Robert Nozick's Anarchy,
State, and Utopia (G. A. Cohen 1995, 4). The metaphor `roused from his
dogmatic slumbers' is, of course, borrowed from Kant, who used these
words to describe the eect on him of reading Hume's Treatise; that is,
Hume made him actively rethink and reassess his common, unreective
assumptions. Thus what Cohen wishes to say is, rst of all, that Nozick's
eect on him was to force him to rethink and reevaluate all his assump-
tions. But this is not all: He also wishes to convey that Nozick's eect on
him was like Hume's eect on Kant. This association with `roused from
his dogmatic slumbers' is not a feature of being awakened from one's
dogmatic slumbers; rather it is a feature dependent on Kant's use of the
(metaphorical) phrase, a historical fact about the metaphorical interpre-
tation of the original token of the phrase, which is now associated with
the later metaphorical interpretation of (the later token of ) the phrase.
I could go on, but these examples should suce to convince you that
there are any number of relations by which, or sources from which, a
property can come to be a candidate for inclusion within a metaphorical
interpretation. This will be no surprise to students of metaphor. Glancing
through the history of theories of metaphor, one is struck by the variety of
principles that have been proposed at one time or another as the so-called
grounds of metaphorical interpretation: relations of similarity, analogy,
dissimilarity, connotational features, iconic features, semantic elds, fea-
tures of ``interaction,'' and on and on. In part the explanation for this
diversity is that dierent theorists have focused on dierent classes of
examples, leading them to dierent grounds to explain their respective
data.9 However, once we see the wide variety of grounds, it should also be
clear that if we want to give a general characterization of metaphor, it
should not be in terms of the specic content of any one of its grounds or
their genealogy.
Traditional theories of metaphor sought to explain the phenomenon by
spelling out a procedure that would yield a literal analysis of each meta-
phor based on one or another ground, something like an algorithm that
would yield, for each metaphor, a synonymous nonmetaphorical expres-
sion. Hence, if they took the diversity of grounds to heart, these theories
would have had to trim their explanatory sails and conne their anal-
yses to specic classes of metaphors, thus surrendering their generality.
And this conclusion would have been fatal for them since it would be
tantamount to denying that the many examples of metaphor and their
Knowledge of Metaphor 113

respective grounds constitute a single unied, or cohesive, linguistic

phenomenon.
Not so for the present theory, whose aim is not to yield any such pro-
cedure or rule but to articulate the semantic structure of a metaphorthe
form that determines its propositional content (in each context) as a
function of certain contextual parameters. From this perspective, the
multiplicity of ``grounds'' poses no theoretical problem. What makes
metaphor a single linguistic phenomenon is that one formal semantic re-
lation governs the context-dependence of its interpretation; the multiple
grounds that underlie the interpretations of dierent metaphors simply
illustrate the variety of ways in which the relevant contextual parameter
can be realized. In what follows, I shall therefore say that a feature related
in any one of these ways to the expression F interpreted metaphorically,
the literal vehicle, is m-associated with F. So, stereotypical, normal,
``metaphor-historical,'' and the exemplied features I shall discuss at
length in chapter 5 are all m-associated with F and are all known (given
my earlier comments on normal features) extralinguistically.
Note that there is no circle here: To be m-associated with F is to be
related to F as a stereotypical feature of Fs, or normal feature, or exem-
plied feature, and so on. None of these relations is itself metaphorical or
linguistic; neither are we analyzing metaphor in terms of these extra-
linguistic relations. They simply register the various sources or relations
by and from which speakers elicit features that belong to the contextual
parameter for metaphor. Although I wish to be conservative in my con-
ception of semantics, I also want to be as liberal, ecumenical, and plural-
istic as possible with these extralinguistic pragmatic notionswith the
range of features that count as m-associated with the literal vehicle.
How, then, to describe the parameter on which metaphorical inter-
pretations depend? For the indexicals, certain features of the world (the
speaker for `I', the time for `now', etc.), determine not only the truth-
values (or references) of what they (or their utterances) say but also their
content. Insofar as they bear on truth-value, those features are function-
ing as (part of ) circumstances of evaluation; insofar as they bear on de-
termination of their content, they function as (part of ) their context of
utterance (or, if you will, interpretation). An equally important, though
not as easily demarcated, fact or feature of the world that plays the same
dual set of roles is the existence of certain background beliefs and atti-
tudes against which an utterance takes place. Following Robert Stal-
naker, I'll call the shared background necessary for the interpretation of
an utterance the set of presuppositions of its context (or context set).10 My
114 Chapter 4

proposal is that the contextual parameter on which a metaphorical inter-

pretation depends is a (subset of the) context set of presuppositions.
This general notion of presupposition is based on the intuitive distinc-
tion between what is asserted, or said, by an utterance (roughly, what we
have called its content in a context) and what it presupposes (in the con-
text). An utterance of `the prime minister of Israel is delighted with the
election of Yassar Arafat' presupposes both that there is a prime minister
of Israel and that Arafat has been elected; it asserts, or says, that the
former is delighted with the latter. A standard test to distinguish what is
asserted or said by a sentence from what it presupposes is to negate the
sentence, for example, `The prime minister of Israel is not delighted with
the election of Yassar Arafat'; what is asserted is obviously dierent from
its armative counterpart, but the presuppositions are the same. Thus the
presuppositions of an utterance are those propositions that can be inferred
from both it and its denial. An analogous test uses a modal variant: `The
prime minister of Israel might have been delighted with the election of
Yassar Arafat' also preserves presuppositions but alters content. If those
propositions were not presupposed, the armative, negative, and modal
utterances would all be inappropriate things to say, either (depending on
one's theory) because they would be false or neither true nor false.
A similar distinction can be drawn between the asserted content of a
metaphor and its ``metaphorically relevant'' presuppositionsthose
properties that must be presupposed to be m-associated with the literal
vehicle because, in their absence, the utterance to be interpreted meta-
phorically would be metaphorically uninterpretable and therefore inap-
propriate. Suppose Romeo utters
(7) Juliet is the sun
in a context c (given the identication of `is the sun' as a metaphor) to
express the proposition that Juliet is, say, the thing around which his life
revolves, that she nourishes him, and that he worships her. This proposi-
tion may, of course, be true or false (depending on the facts about Juliet's
and Romeo's relation to each other), but, for it to have one or the other
truth-value, these properties (perhaps among others) must be presupposed
to be m-associated with the word `is the sun'. These presuppositions are
preserved when we negate (7) (in the same context and where the predi-
cate is still interpreted metaphorically):
(7a) Juliet is not the sun,
meaning that she is not the person around whom Romeo's life revolves,
or that she does not nourish him or that he does not worship her. Like-
Knowledge of Metaphor 115

wise, the same properties are presupposed when we utter (in the same
context, with the same interpretation)
(7b) Juliet might be the sun.
If those properties were not presupposed to be m-associated with the lit-
eral vehicle, `is the sun', none of these three utterances would be successful
or appropriate under this interpretation. Of course, dierent properties
might be presupposed to be m-associated with `is the sun', in which case
(everything else being equal) the utterance would then have a dierent
metaphorical interpretation whose appropriateness, or interpretability,
would also depend on those presuppositions being held. But unless the
expression to be interpreted metaphorically is presupposed to be m-
associated with some such identied set of properties, it won't be know-
able at all in the context under what conditions it is true, and hence
appropriate.
Now, if the contextual parameter for metaphor is a set of ``metaphori-
cally relevant'' presuppositions, the character of an expression F inter-
preted metaphorically, or (using the term of art I introduced earlier) the
character of the metaphorical expression `Mthat[F]', is a function from
the ``metaphorically relevant'' set(s) of properties presupposed to be m-
associated with F in its containing sentence S in the context c to a set of
properties P.11 In other words, corresponding to the rule (Dthat) for
dthat-descriptions, we can propose an analogous rule of character for
metaphorical expressions:
(Mthat) For every context c and for every expression F, an occurrence
of `Mthat[F]' in a sentence S ( . . . Mthat[F] . . .) in c (directly)
expresses a set of properties P presupposed to be m-associated with F in
c such that the proposition h. . . P . . .i is either true or false in the
circumstance of c.
(Mthat), like (Dthat) and (I), is a rule of character that a speaker knows
simply in virtue of his knowledge of language. As we shall see, in terms
of it we can spell out the constraints on metaphorical interpretation de-
scribed in chapter 2. But because it is a semantic rule, it also doesn't make
explicit the pragmatic, extralinguistic notions that describe the context;
for example, what it is to be m-associated with F, what it is to be pre-
supposed, and exactly what xes Pwhat is, as it were, the mechanism
encapsulated in the expression-relation. In the coming sections I shall
attempt to explain some of these notions in more detail: (i) the attitude
of presupposition that is appropriate for metaphor, and how it diers
from the simple attitude of belief; (ii) the sense in which metaphorically
116 Chapter 4

relevant presuppositions are presuppositions; (iii) the relation between

speakers' presuppositions tout cour and the specic presuppositions that
are needed for the interpretation of a metaphor; and (iv) the degree to
which the mechanism that xes the content of the metaphor in a context,
the set of properties P, is semantic. In chapter 5 I shall discuss at length
one particular class of metaphors based on the extralinguistic perception
of exemplication-relations. But before continuing, let me add two ex-
planatory glosses.
First, according to (Mthat) the content of a metaphorical expression in
a context, the property(ies) P, is determined by two factors: the function
that is the character of the metaphor and the context set of presupposi-
tions that is the domain of the function. In setting up the semantics, we
therefore have a certain amount of freedom in dividing the labor. At one
extreme, we can dene the domain of presuppositions as narrowly as we
wish, eliminating all but those that actually belong to P, and make the
function an identity-mapping. At the other extreme, we can be liberal
with regard to membership in the context set of presuppositions, and let
the function do all the sorting and selecting of P. In what follows, I tend
toward the rst strategy. Although it is not yet entirely clear how, and
how much, these two ways of cutting up the pie actually make a dier-
ence, I have tried to keep the semantics simple and leave messy details to
the pragmatics or description of the context. However, one can imagine
other ways of yielding the same results.
Second, the content (in a context) of the metaphorical expression is a
set of properties such that the proposition expressed by its whole utter-
ance is determinately true or false. But, to anticipate a possible objection,
I should emphasize that this content of the metaphor in a context is not,
or not always, its full or sole interpretation, or meaning, or signicance.
As we'll see at length in chapter 7, the characters of many metaphors are
equally important for the information or signicance they convey; fur-
thermore, not all cognitive signicance carried by the character of a
metaphor can be expressed propositionally, let alone as part of its
propositional content in its context of utterance. As I'll also argue, this
character-istic information of the metaphor is generally manifest in a va-
riety of ways other than knowledge of truth-conditions; hence it is best
distinguished from the truth-conditions of the metaphor, or its content in
a context. At the same time, it is also not always evident precisely where
to draw the line between those components of its cognitive signicance
that are relevant to the truth-evaluation of the metaphor and those that
are not. Without trying to draw sharp boundaries around truth-conditions
Knowledge of Metaphor 117

(or content), I'll return to this problem in chapter 5, section III to illus-
trate how this issue bears on criteria of individuation of metaphors.

II Presupposition, Belief, Pretense, and Presupposition

The pragmatic notion of presupposition I am employing was rst pro-

posed as an alternative to the semantic notion that was popular among
philosophers and linguists from the 1950s through the 1970s. According
to the semantic notion, presupposition is a relation between two proposi-
tions P and Q such that P (semantically) presupposes Q i both P neces-
sitates Q and not-P necessitates Q, where P necessitates Q i whenever
P is true, Q is also true. (Where Q is false, P is claimed to be truth-
valueless.)12 According to the pragmatic notion proered by Stalnaker
and others, we explain the linguistic presuppositions of sentences not in
terms of semantic relations between their propositional contents, but by
features of their use or properties of their users. The pragmatic notion is
guided by two central ideas. The rst is that presupposition is a psycho-
logical attitude like belief (whose object is, depending on one's theory,
either a proposition or sentence). The second is that a central function of
presuppositions is to constitute the context on which the appropriateness
(including interpretability) of an utterance depends. That is, unlike the
semantic notion that takes presuppositions to be required for truth-
valuedness, the pragmatic notion requires them for appropriateness. Fur-
thermore, the presuppositions of a sentence that are required for its
appropriateness are ultimately to be dened in terms of the presuppos-
itions that a speaker must make for his utterance of the sentence to be
appropriate.13
Among the attitudes, what distinguishes presupposition is that the
subject takes the object of the attitude (be it a proposition or sentence) to
be ``common knowledge'' in the context, part of a doxastic background
mutually shared by the other participants. As a rst approximation, we
might therefore dene presupposition as follows:
(P1) A proposition p is a presupposition of a speaker S in a context c
(or S presupposes p in c) i (1) S believes that p; (2) S believes that the
other members of c believe that p; and (3) S believes that the other
members of c recognize that he (S) believes that p.
(P1) captures the relevant kind of agreement for presupposition: agree-
ment among the members of the context about p rather than agreement of
p with the world. Thus presuppositions are the ``common knowledge'' of
118 Chapter 4

the participants despite the fact that they need not be true, in which case
they cannot, strictly speaking, be knowledge. However, even (P1) is too
strong because the presupposition need not even be believed by S and by
the other conversational participants. Sometimes, of course, it is believed,
but, as we saw with the m-associated stereotypical features (e.g., (3)), at
times the presuppositions are denitely known to be false. In yet other
cases, the participants may represent themselves as believing propositions
they in fact do not believe, adopting the belief only for the sake of the
discourse and within its local connes. In some of these cases, Stalnaker
suggests that the speaker pretends to believe the presupposition; she acts,
or makes utterances, in accordance with but without actually believing
the presupposition.14 Without dwelling on all the subtle dierences be-
tween these cases, I shall suppose that what is necessary for presup-
position is neither belief nor even pretending to believe but that the
participants represent themselves as if they believe that p. So, of the vari-
ous conditions in (P1), only the third is perhaps necessary: S must believe
that the other contextual participants can recognize the presuppositions
she represents herself as having, whether or not either she or they actually
believe them or she believes that her audience believes them.
It should also be mentioned that even if presupposition were pretending
to believe, it would not follow that the contents of metaphorical assertions
``involve the pretense that something is the case when it is not.''15 The
speaker's attitude toward his presuppositions (or the presuppositions of
his utterance) should be distinguished from his attitude toward the
asserted content of his utterance. Even if we pretend that elephants are
delicate, dainty creatures when we know that they are not, I can use the
sentence `The sumo wrestler is an elephant' to assert (under a metaphori-
cal interpretation) with no pretense that the wrestler is delicate and
dainty.
Let me turn now to the second clause of (P1), which seems to require
that the speaker believe that her fellow contextual participants believe the
presuppositions at the time of utterance. This is also not in general true
and not true in particular for the presuppositions of a metaphor. Let me
rst discuss a nonmetaphorical case. Often a speaker chooses to convey
some piece of information to her audience (because she knows they do not
share it), not by asserting it, but by asserting something else that presup-
poses that information. For example, I let you know that Bill in whom
you have taken a sudden romantic interest is married by saying: ``And let
me introduce you to Bill's wife, too.'' The reason for this indirect form of
communication might be either tact or, because it forces the hearer
Knowledge of Metaphor 119

actively to draw out the presupposition, eectiveness. However, from

cases like this we should not conclude that presupposition is pretended
rather than actual belief. What the example shows is that presuppositions
need not be presupposed: they need not be held in common in advance of
or initial to the time of the utterance that presupposes them. Instead a
speaker may come to make a presupposition only after, or as a conse-
quence of the fact that, she utters a sentence expressing a proposition
whose appropriateness is contingent on the presupposition. (I'll return
in the next section to sentence-presupposition and ``making presupposi-
tions.'') That is, the only sense in which a presupposition p is pre-, or prior
to, its presupposing assertion q is that it is logically prior in its respective
context; that is, p is a condition on the appropriate assertion, or inter-
pretability, of q. We can make this more precise in terms of a model of
discourse.
Each contextual participant brings her own individual context set of
presuppositions to a conversation but, because conversation demands
cooperation, individual context sets of dierent participants typically
coincide, at least suciently to enable conversation to proceed. Let the
intersection of the individual context sets of the participants be the shared
context set of the conversation. A context is ideal just in case the individ-
ual context sets of its participants are each identical with the shared con-
text set.16 Of course, given our dierent contingent doxastic histories, the
ideal is rarely realized, especially if each individual's context set includes
all her dispositional beliefs. A context is close enough to ideal if and only if
the dierences that exist do not disrupt the business of the contextdo
not render utterances inappropriate or uninterpretable and do not block
communication or shared understanding. When individual context sets of
participants do diverge suciently from the shared context setsu-
ciently to aect the interpretation or appropriateness of an utterance
there is a breakdown in communication or conversation.
Think of a discourse over time as a series of successive stages, each of
which is dened by the shared context set of presuppositions (csi ) of its
participants at that time. At any discourse stage n, the utterance of a
sentence s can aect csn in two ways. First, if the utterance of s at n asserts
a proposition p that does not belong to csn , then p is added to csn , yielding
what is presupposed at the next stage, csn1 . But secondand more
important for our purposesthe utterance of s will also in general cause
its participants to infer various other propositions in addition to p, the
proposition it asserts. Among these will be the proposition that the
speaker is speaking and any other information that the speaker believes
120 Chapter 4

his audience must infer from his utterance in order to interpret and
evaluate it as an appropriate assertion in the context set csn . These in-
ferred propositions might include ones that are not already occurrently
believed in common by all the contextual participants, propositions that
belong to the speaker's individual context set but not to the others'. By
making a statement that rests on them, the speaker reveals that he
believes that the others will add them without challenge to their own in-
dividual context sets and that he intends for them to do so.17 All of this
information is presupposed as part of csn inasmuch as it is necessary in
order for the speaker to express p using s at csn even though it comes to
be held only after the speaker opens his mouth and performs the utter-
ance. In these cases, we will say that by uttering s the speaker makes
whatever presuppositions are necessary in order for s to express p. And
inasmuch as the presuppositions belong to the context set csn relative to
which p is asserted, they are prior to the assertion of p for which they are
presupposed.18
The same story holds for metaphor. The metaphorically relevant pre-
suppositions need not be held temporally before the speaker utters the
metaphorical expression or sentence, prior to the speech act. Both the
proposition that the speaker is uttering a metaphor at n and all the met-
aphorically relevant presuppositions that turn out to be grounds for its
interpretation are counted as part of the context set csn , even if they come
to be articulated or held only after the utterance event.
An especially important class of presuppositions of this sort are the
properties associated with so-called creative metaphors, those to which
Max Black referred when, in the course of criticizing comparison theories
of metaphor, he made his now-famous statement that:
It would be more illuminating in some of these cases to say that the metaphor
creates the similarity than to say that it formulates some similarity antecedently
existing. (1962, 37)
Black has since defended this claim as an ontological thesis,19 but these
creative metaphors may appear to raise a problem for our account even if
we understand ``creative'' in an epistemic sense. On that reading, a meta-
phor creates a similarity or property just in case it makes its speaker-
hearer aware of a similarity or property he, and possibly everyone else,
had not previously perceived. Metaphorical creativity of this sort is closer
to the way Hollywood producers create starlets than the way God created
the heavens and earth.20
Let's grant that some metaphorsespecially in scientic and literary
contextsaccomplish this intellectual feat. Someone might now object:
Knowledge of Metaphor 121

``If, as you claim, the interpretation of a metaphor depends on con-

textually presupposed properties, any property expressed by a metaphor
must occur in some proposition entailed by the speaker's context set of
presuppositions; therefore it must be presupposed prior to the metaphor.
But then no metaphor can ever be creative or innovative. No metaphor
can express a property that its speaker-hearer discovers only by its utter-
ance; by hypothesis such a property will not belong to the context set
from which the content of the metaphor is drawn. Therefore, your theory
only accounts for the least interesting metaphors, those based on what
Black called a ``system of associated commonplaces.'' It cannot account
for the ``vital'' interpretations that originate with metaphorical expres-
sions themselves.''
My response to this objection should be clear by now. The context set
relative to which the content of the metaphor is asserted consists of all
information inferred from the utterance that is necessary for its inter-
pretation; therefore, it will include any proposition or property that is
required for its interpretation even if the speaker comes to recognize the
propertyeven if it is ``created''as an outcome of the utterance. These
properties may be products of his imaginative eorts as he attempts to
construct an interpretation of the metaphorical sentence, or they may be
features he notices from the syntax or sounds of the words used. If, as I'll
argue in the next section, it is the interpreter of a metaphor who makes
the presuppositions necessary for its interpretation, we might also say that
the metaphor makes the novel properties or similaritiesthat is, makes us
see or notice them.21 Indeed the speaker himself may not ``see'' these
features until after he has uttered the metaphoror only by uttering it
and therefore he will not have presupposed them prior to the utterance.
Perhaps, then, the speaker, prior to the utterance, also could not have in-
tended to express the content that depends on these ``created'' features.22
None of these properties should therefore count as ``antecedently given
commonplaces'' in Black's sense; yet they count as full-edged pre-
suppositions in our sense inasmuch as they belong to the context set csn
conditional on which the content of the metaphorical sentence can be
asserted at the nth stageregardless of the real time relation between the
utterance and presuppositions or the temporal order in which their actual
processing proceeds.

III Utterance-presuppositions and Metaphor

(P1) characterizes presupposition in terms of ``common knowledge'' or a

shared doxastic background. But it should be common knowledge (at
122 Chapter 4

least among readers of this book) that not all common knowledge is pre-
supposed as part of the context of each particular utterance. There are
obviously an innumerably large number of beliefs that are ``obviously
true, and we each recognize that the other knows that [they are] obviously
true'';23 in one sense of the term, these are ``presuppositions'' we share. But
if we want to capture the second idea that underlies the pragmatic notion
of presupposition, the idea that a primary function of the presuppositions
of an utterance is to demarcate the boundaries within which the utterance
is acceptablewhich, for a metaphor, would include the presuppositions
that enable it to be both interpretable and interpreted with a particular
contentthen we need some way to limit the context set to the specic
body of presuppositions that facilitate and constrain the understanding
and appropriateness of the given utterance. For this purpose it is not
enough to say that presupposition is an utterer's attitude; we need criteria
for what it is for an utterance to have (or make) presuppositions.
Unfortunately, at this point in our story, there is a complication, largely
owing to a disanalogy between the general relation of an utterance to its
presuppositions and the relation of a metaphor to its presuppositions.
Nonetheless I shall argue that there remains enough of a relation to justify
thinking of the latter on a model of the former. However, to put the dif-
ferences between them in perspective, let me begin with the general rela-
tion between an utterance and its presuppositions and then contrast it
with the case of metaphor.
To begin with, let's make explicit the relation between a speaker's pre-
suppositions (as in (P1)) and her utterance.
(P1*) Speaker S presupposes a proposition p in a context c by uttering
the sentence s i (1) S represents herself as believing that p; (2) S
represents herself as believing that the other members of c represent
themselves as believing that p; and (3) S represents herself as believing
that the other members of c recognize that she (S) represents herself as
believing that p from her utterance of s.24
The last condition of (P1*) requires that the presuppositions of the
conversational participants somehow be manifest in the utterance. This
condition was the main insight underlying the semantic notion of pre-
supposition, although its criterion that the presuppositions be logically
entailed (by the sentence and its negation) obscured the variety of syn-
tactic, semantic, and pragmatic sources of presuppositions. Therefore,
instead of saying (pace the semantic notion) that the asserted sentence is
false (or truth-valueless) where the presuppositions fail, we shall say that
Knowledge of Metaphor 123

the utterance is, more broadly, inappropriate. Let's also say that we can
reasonably infer that p from an utterance u of a sentence s by S if and only
if it would not be appropriate for S to utter u unless S presupposes p. We
can now state what it is for an utterance u of a sentence s to presuppose p
or to require p as a presupposition:
(P2) An utterance u of a sentence s presupposes that p (or requires p as
a presupposition) i we can reasonably infer that p from u.
Likewise, we can dene sentence presupposition in terms of utterance
presupposition:
(P3) A sentence s presupposes that p i all (or all normal) utterances of
s presuppose that p.
Finally, in terms of (P2) we can dene a narrower conception of the
context set of presuppositions for an utterance:
(P4) The context set of presuppositions for an utterance u are all and
only those presuppositions required for (or presupposed by) u.
As I said in section II, all and only those propositions that can be rea-
sonably inferred belong to the context set of an utterance (including those
necessary for its interpretation) whether or not they were commonly held
by the participants before the time of the utterance.
Among the presuppositions required for the appropriateness of utter-
ances, philosophers and linguists have given their greatest attention to
those invariantly determined by their respective sentences, and, in partic-
ular, by conventional features of their form and content. They have con-
centrated on the presuppositions of, for example, lexical items (e.g., words
like `even', `only', `again', the determiner `the'), syntactic congurations
(e.g., cleft and pseudo cleft constructions), and compound sentences as
they are computable from those of their sentential components (the
so-called projection problem). I shall assume that the best theory that
explains and predicts these presupposition phenomena as they arise for
the literal use of language carries over to metaphor. However, there is one
prima facie dierence between the presuppositions of a metaphor and the
presuppositions generated by the conventional features of sentences.
Take our earlier example, ``And let me also introduce you to Bill's
wife,'' which presupposes that Bill is married. Even if it is not already
presupposed by the hearer at the time of utterance, this presupposition
will be reasonably inferred by any contextual participant in virtue of the
form and meaning of the expression `Bill's wife'. He'll reason as follows.
``Since it would be inappropriate (for some reason) for any speaker nor-
124 Chapter 4

mally to utter that sentence unless he held that presupposition, the

speaker must himself believe it and assume that I (the hearer) either
already know it or will accept it without challenge. Therefore, I hereby
add the proposition that Bill is married to my presuppositions.''25 The
``would'' in this piece of reasoning is signicant. These presuppositions
are nomically related to their respective utterances: They must be held by
any (normal) member of the linguistic community if any (normal) utter-
ance of the presupposing sentence is to be appropriate in its context.
Given his general intention to belong to the linguistic community and to
be speaking English, the speaker makes presuppositions simply by uttering
sentences that require them. Just as the words `Bill's wife' mean what they
do regardless of their speaker's occurrent intentions on a given occasion
(but given his general intention to be speaking English, etc.), so the
speaker's presuppositions are ``read o '' his utterance without our exam-
ining his own individual occurrent beliefs and intentionsalthough we
assume that, as a competent sincere speaker, he actually holds the pre-
suppositions his utterances require.
Indeed, whether or not a speaker (occurrently) believes the pre-
suppositions of his utterances, he is committed to themand can be held
responsible for themjust as he is to their meanings. As Stalnaker writes:
``[T]he act of making a presupposition, like the act of meaning something,
is not a mental act which can be separated by an act of will from overt
linguistic behavior.''26 The ``overt linguistic behavior,'' the utterance, is
not simply evidence that the speaker presupposes what the utterance
requires to be appropriate; to utter a sentence that requires them is to
make its presuppositions. Just as meaning (or expressing a proposition)
is not an act distinct from uttering a sentence that has that meaning
(or expresses that proposition), so presupposing, or making a presup-
position, is not an act distinct from using a sentence that requires the
presupposition.27
The relation between a metaphor and its presuppositions is prima facie
disanalogous to this nomic relation that holds for conventionally deter-
mined presuppositions. There is no nomic relation between an arbitrary
metaphor (individuated by its linguistic form or character) and a particu-
lar set of presuppositions. For each metaphorical expression (type) there
is no one set of presuppositions that remains invariant across contexts or
that must belong to any context set in which an occurrence of a token of
the metaphor (type) receives an appropriate interpretation. Dierent sets
of presuppositions may each be sucient to determine a (possibly dier-
ent) interpretation either for dierent tokens of one metaphorical expres-
Knowledge of Metaphor 125

sion (type) or even for a single token; no single context set is necessary for
any utterance of the metaphor to have an interpretation or an appropriate
interpretation. There is also no train of reasoning analogous to that
rehearsed in the previous paragraph whereby an interpreter who does not
already share the speaker's particular set of presuppositions would be
drawn to conclude, on grounds of what would normally be presupposed
by utterances of the metaphor, that he must adopt a particular presup-
position set. Therefore, no speaker makes particular metaphorically rele-
vant presuppositions simply in virtue of his utterance of a metaphor, in
the way in which he does make their respective presuppositions simply by
using sentences with certain conventional forms and meanings.
These prima facie dierences are not insignicant. Nonetheless I think
we can narrow the gap between the presuppositions of a metaphor and
those of general utterances, even if we cannot close it. On the one hand,
not all instances of literal language determine particular required pre-
suppositions for their appropriate utterance. On the other hand, and
contrary to rst appearances, some metaphors do have something like a
normal required presupposition, at least as their default interpretation.
Finally, the same kinds of rules that govern the appropriateness of utter-
ances in general relative to their presuppositions also govern metaphor.
So, although metaphor may be underdetermined relative to its presup-
positions (especially compared with the general nonmetaphorical case),
these three considerations seem to me sucient to justify my proposal
that the contextual parameter for metaphorical interpretation is a context
set of presuppositions. I'll review these considerations in order.
First, metaphorical expressions (types) are not unique in requiring a
presupposition but no unique or specic one; the same is true of some
literal expressions (types). Only after their content on a particular occa-
sion of utterance is determined do they acquire a specic presupposition.
Not surprisingly, demonstratives are one example. Any utterance of the
sentence `. . . He[D] . . .' requires a presupposition, namely, that there is a
unique male demonstratum, without presupposing the existence of a par-
ticular thing presented by D in its context. Only when we x the content
of the complete demonstrative in its context can we identify its partic-
ular presupposition. Likewise, only when we x a particular content for
the metaphor in its context can we identify its particular context set
of presuppositions. In this respect, metaphors are no worse o than
demonstratives.
Second, contrary to the common impression, some metaphorical
expressions do have particular metaphorically relevant presupposition
126 Chapter 4

sets they normally presuppose across contexts, that is, in all context sets
relative to which they are normally appropriate. As we'll see in chapter 8,
routinized (but not yet dead) metaphors are expressions whose inter-
pretations are unquestionably metaphorical (i.e., their content in a con-
text is computed as the value of applying their character to a context
set of presuppositions) but they employ, or require, the same presupposi-
tions on all (or on all normal) occasions in which they are used. The
same is true of metaphorical expressions whose interpretations rest on
m-associated features that are stereotypical (e.g., metaphors whose literal
vehicles are natural kind terms, such as `is a gorilla') or normal (e.g.,
color terms). Neither of these metaphorical interpretations is context-
independent, but they could be called (relatively) context-invariant be-
cause they carry their presuppositions with them from context to context.
And because they are more easily ``accessed'' than more context-specic
presuppositions (like those involving the exemplied features I'll discuss
in ch. 5), these presuppositions may furnish ``default'' interpretations
when others are not, or are not yet, available.
Of course, at this point one might object that these ``normal'' meta-
phorical interpretations are, precisely because they are context-invariant,
intuitively dierent from ``extranormal'' metaphorical interpretations,
say, in the way in which poetic metaphors (or poetic metaphorical inter-
pretations) intuitively dier from nonpoetic ones. And, therefore, they are
not good examples. Yes and no. Routinized metaphors with their normal
interpretations are, I agree, not good examples of the relation between
poetic, or highly context-specic, metaphors and their presuppositions;
between the presuppositions of poetic metaphors and the ordinary case of
required presuppositions of an utterance there remains a large gap. But
insofar as routinized metaphors, which are no less metaphorical for being
routinized, may nonetheless serve as examples of how a metaphor can
normally require a presupposition, they show that the source of the dif-
ference may lie with the specic kinds of presuppositions employed in the
interpretations of some metaphors, not with the general structure of pre-
supposition that is needed for metaphorical interpretation.
The third consideration that should be taken into account in judging
whether the m-associated features of metaphorical interpretations func-
tion like the presuppositions required for the general appropriateness of
an utterance harks back to Stalnaker's guiding idea, that a primary func-
tion of the presupposition set for an utterance is to constrain and enable
what the utterance can appropriately assert. Given the structure of a
context set of presuppositions, Stalnaker both articulates an abstract
Knowledge of Metaphor 127

conception of communication as a rational activity and, in turn, uses it to

justify certain principles or rules of interpretation and communication.
The abstract structure has this form: The presuppositions demarcate a
range of possibilities (worlds) among which the agent's assertionshis
informative utterancespartition those that are actual, those that remain
possibilities for the future, and those ruled out as impossible. Assuming
that the agent's set of presuppositions is consistent and deductively closed,
utterances that express propositions that are either inconsistent with the
presuppositions (hence false in all circumstances compatible with the
presuppositions) or entailed by them (hence true in all such circum-
stances) are therefore not informative and hence not assertions: They
make no partitions. These statements are inappropriate, at least insofar as
they purport to be assertions. But this is not the end of the story. Where
an utterance is apparently inappropriate, given this structure, we can
draw any of a number of conclusions. Either the presuppositions are not
what we have supposed them to be, or the speaker should be interpreted
as asserting something else, or indeed there has been a violation that
makes the utterance deviant unless we can otherwise explain it, using a
Gricean maxim, in terms of some other implicature.
Do the presuppositions of a metaphor have a similar rational structure
that constrains its interpretability? To begin with, we should keep in mind
that there are two sets of presuppositions at work in the interpretation
and evaluation of a metaphor. Or more preciselyalthough I shall con-
tinue to speak of them as if they were two setsthe presuppositions
function in two roles. The rst set consists in the presuppositions relative
to which we evaluate the appropriateness in the context of the proposi-
tion that is asserted by the utterance under its metaphorical interpre-
tation. The second set consists in presuppositions relative to which we
generate an appropriate metaphorical interpretation for the string. Call
the rst set the A(ssertion)-set of presuppositions; call the second set the
I(nterpretation)-set. The A- and I-sets can be mutually inconsistent; for
example, some m-associated feature presupposed in the second set may be
known by the speaker to be falseand therefore not something assert-
ablerelative to the rst set of presuppositions. (In other cases, where
one presupposition plays two roles, it can belong to both sets.) In general,
then, the two sets of presuppositions are insulated from each other. Yet,
the A-set can indirectly aect the interpretation of a metaphor and hence
the I-set. So, consider two rules of interpretation for metaphor (modeled
after Gricean rules of rational cooperative linguistic behavior):
128 Chapter 4

(Inconsistency) Do not interpret someone as to make his sentences

express some proposition inconsistent with his presuppositions.
(Redundancy) Do not interpret someone as to make his sentences
express some proposition already entailed by his presuppositions.
The relevant presuppositions for both of these rules belong to the
speaker's A-set. However, an analogous rule applies to the I-set:
(Presuppositional Inconsistency) Do not interpret someone so that the
content of his assertion requires presuppositions that are either
inconsistent with his other I-presuppositions at that time or that are
internally inconsistent.
Note that all these rules say ``do not'' rather than ``one cannot'': once we
establish rules or maxims of interpretation or appropriateness, speakers
tend to exploit those very rules by acting as if what is presupposed is
not (representing themselves dierently) and thereby overruling them.28
However, if we listen closely to subtle dierences between interpretations,
it is possible to become sensitive to nuances whereby we can detect where
an interpretation follows a rule governing the context set of presupposi-
tions and where it arises from the violation of a rule by overriding it.
For example, consider the following instance of (one type of ) mixed
metaphor, composed from two lines, the rst by Homer (to express the
subject's ability to inspire terror, fear, and the foreboding of the un-
known), the second by Byron (to express its subject's gracefulness and
elegance):
*(8) Apollo came like night, while Venus walked in beauty, like night.29
Or consider the following `sun' metaphors that we discussed briey at the
end of chapter 2, each of whose (explicit or anaphoric) occurrences of the
metaphorical expression makes dierent, inconsistent presuppositions
about the features m-associated with the literal vehicle `is the sun':
*(9) Juliet is the sun, and Achilles is, too,
(or the even less acceptable)
*(10) Juliet is the sun and Achilles is the sun.
Or imagine the following scenario. A corporate executive impatiently tells
his sta:
(11a) Would someone please come up with a green thought.
The sta members scurry to dash out ideas on their notepads but after
handing them over, the boss complains:
Knowledge of Metaphor 129

(11b) You always give me your green thoughts before you have worked
out any of the details.
In (11a) `green' is interpreted relative to a presupposition set in which the
predicate is contrasted with the color terms `yellow' or `brown' and
expresses the properties of being fresh, novel, or original; in (11b) it is
interpreted relative to a context set in which `green' is contrasted with a
term for a ripe, mature, or developed (say) fruit, like `red' or `(bright)
yellow'. In the rst, the predicate has positive evaluative force, in the
second, a distinctly negative one.30
Or consider the classic mixed metaphor, supposedly produced by
Ronald Reagan:
(12) The ship of state is sailing the wrong way down a one-way street.
There is something unacceptable about (12), but it is not obvious what.
Not all mixed metaphors, or sentences containing multiple metaphors, are
equally unacceptable; some indeed are not only powerful but appear to be
powerful because they are mixed: thus
(13) To take arms against a sea of troubles.
What is the dierence between (12) and (13)? To begin with, (12) is am-
biguous; we can mark the two interpretations in terms of the dierent
scopes of the Mthat-operator (as I shall show further in the next section):
(12a) Mthat[`the ship of state is sailing'] Mthat[`the wrong way down a
one-way street'].
(12b) Mthat[`The ship of state is sailing the wrong way'] Mthat[`down a
one-way street'].
The interpretation of (12a) is that the government is following a path that
will inevitably lead to some sort of confrontation or collision; the inter-
pretation of (12b) is that the government is following the wrong policies in
an uncorrectable or irreversible manner, in a way that prevents it from
changing its direction or turning back. Suppose now that we resolve this
ambiguity in context and decide on one or another interpretation. Al-
though there is no diculty determining its content, (12) sounds decidedly
badand much worse than (13). Why? One explanation might be along
the lines of our present discussion. To interpret the prepositional phrase in
(12), `down a one-way street', we must ll in the context set with the pre-
supposition that the state is an automobile or car. But this presupposition
is inconsistent with the presupposition required for the interpretation of
the rst part of (12), namely, that the state is a ship.31 Hence the inter-
130 Chapter 4

pretation of the utterance requires that we make inconsistent presup-

positions. In (13), on the other hand, although the two metaphors are
dierent and indeed independent of each other``independent'' in that
the order in which they are respectively interpreted makes no dierence
their presuppositions are not inconsistent. In fact, they oddly comple-
ment, and strengthen, each other. Let me briey explain.
Although it is always possible to ask why an author chooses to use one
particular metaphor rather than another, with mixed metaphors where
she abandons one for another midcourse through the utterance, the
question of choice is inevitable. Moreover, where the dierent metaphors
could express the same or equivalent contents, an answer must appeal to
their characters and the presuppositions associated with their literal vehi-
cles. If we can show how, by shifting metaphors, and the associated pre-
suppositions, the author was able to communicate further information
she could, or would, not have communicated by sticking with the same
metaphor, the mixing will be justied. In (13) this is indeed the case. Al-
though the content of (13) is hto overcome, or struggle against, vast and
manifold troublesi, the character of (13) informs us, in addition, of the
apparent futility of so doing, the prima facie impossibility of taking arms
against seas. This information, communicated precisely by the mixed
metaphorwe might say: category mistakelends (13) much of its
power as a metaphor. In (12), on the other hand, there is no comparable
redeeming information conveyed by the characters of the multiple meta-
phors that would justify the change of metaphor midcourse in one context
of utterance. In fact, the inappropriateness or unacceptability of (12) may
also derive in part from the very fact that it raises without answering the
question of why the author chooses to change metaphors in the middle of
his utterance.
To return now to our main line of argument: In all these examples we
interpret the rst occurrence of the metaphor relative to one presupposi-
tion set, and continue, prima facie in the same context, to interpret an-
other occurrence of the same metaphor (or a metaphor from the same
family or schemaan idea to which I'll return in ch. 5) relative to pre-
suppositions that are inconsistent with the rst set. The result violates the
rule of Presuppositional Inconsistency, thereby generating a mixed met-
aphor. (To repeat my earlier point, this violation does not render the
metaphor absolutely uninterpretable or incoherent; but we can sense a
dierence between the kinds of interpretations that conform to the rules
and those that do not.) Here the rules that govern metaphorical interpre-
tation are the same rules of rational cooperative linguistic behavior that
Knowledge of Metaphor 131

govern the appropriateness of utterances in general. The one dierence is

that metaphor does not typically require a unique set of presuppositions
like those that may be required by sentences. Unlike speakers' absolute
commitments to the presuppositions that depend simply on the conven-
tional form and meaning of sentences, metaphorical interpretation
involves conditional commitments given presuppositions set up at one
stage that transform the context of interpretation at later stages and
thereby constrain the interpretation of subsequent metaphors. In each of
(8) and (9), for example, the utterance of the rst conjunct Pi under a
metaphorical interpretation occurs in a context csi that contains all the
presuppositions Qi that are required for the appropriate interpretation of
Pi . However, the second conjunct Pj in each of the examples is interpreted
relative to csj , which results from adding the asserted content Pi to csi ,
and also contains the further presuppositions Q j required for the inter-
pretation of Pj . If Q j in turn are inconsistent with Qi (also in csj ), either Pj
will be inappropriate in csj or we must insulate csi from csj , allowing the
speaker to shift between disjoint contexts in the course of a single utter-
ance. Insofar as we seek to avoid either of these consequences, our inter-
pretation of one metaphor relative to a given context commits us to
certain presupposition requirements for our further interpretation.32
To sum up, I have tried to indicate both dierences and similarities
between the relations of a metaphorically interpreted utterance to its rel-
evant presuppositions and of appropriate utterances to their (conven-
tional) presuppositions. The dierences are not mere artifacts of our
theoretical apparatus for representing presuppositions; they reect genu-
ine dierences between the role we are asking presuppositions to play in
accounting for metaphorical interpretation and the role they play more
generally in explaining the appropriateness of utterances. Unlike the pre-
suppositions for the general appropriateness of utterances, which can be
identied either by means of logical relations (e.g., entailment) or by
means of syntactic or lexical features of the utterance, the metaphorically
relevant presuppositions for the interpretation of an utterance cannot be
uniquely identied in such a way. Some m-associated features, like
stereotypical or normal features, are plausibly known by each competent
member of the linguistic community and, if only by default, are there for
one to utilize as a metaphorically relevant presupposition. But for meta-
phors that are more sensitive to their respective actual contexts, and
whose interpretations are more specic to their contexts, there are no
general rules or recipes to identify their metaphorically relevant pre-
suppositions. For dierent kinds of metaphors, there are dierent clues or
132 Chapter 4

heuristics we use, the kinds of rules of thumb or start-up techniques that

enable us (to the degree to which we are able to) to perceive similarities
or, as I'll discuss at length in the next chapter, to perceive what is exem-
plied or sampled by something. These skills, however, cannot be reduced
to rules; on the contrary, they are best put into practice enhanced by a
heavy dose of imagination.
Despite these dierences, the three considerations I mentioned later in
this section give us, I propose, sucient reason to treat the m-associated
features that are employed in metaphorical interpretation as contextual
presuppositions, which, in turn, function as the domain on which the
character of a metaphor operates, yielding a content in a context. This
next part of the story, however, is also more complicated, and controver-
sial, than I have so far let on. In the next section, I shall turn to one
question for my claim that the relation between metaphorical character
and the context set of presuppositions is a truly semantic relation on the
model of the relation between an indexical character and its contextual
parameter. My answer to this question will, like the conclusion of this
section, also be less than conclusive, but I hope the complications and
qualications give the reader a better grip on the issues illustrated in the
next chapter.

IV Is Knowledge of Metaphorical Character Really Semantic?

The character of a metaphor is formally dened as a function from con-

text sets of presuppositions to sets of properties. The value of the function
(in a context) must therefore be a determinate truth-conditional or prop-
ositional factor, or at least as determinate a factor as those assigned by
the characters of the indexicals and demonstratives, a factor that will en-
able us to evaluate whether utterances are true or false. But (1) is the
character of a metaphor suciently well constrained to assign it a content
that, when evaluated at a circumstance, yields a truth-value? And (2) if it
is, is the mechanism that is formally represented by the character function
a semantic mechanism, again, like the mechanisms for the indexicals, a
rule that the speaker knows simply in virtue of his semantic or linguistic
knowledge?
The rst question is a skeptical challenge to my project tout cour. In
chapter 1 I said that the appearance is that utterances containing meta-
phors, no dierent from exclusively literal utterances, make assertions
that are judged true or false. But if they are truth-valued, they must have
truth-conditions or content. The burden of argument therefore falls on
Knowledge of Metaphor 133

those who deny that this appearance is reality. Yet there is deep resistance
to the appearance. As Davidson puts it: ``It should make us suspect the
theory that it is so hard to decide, even in the case of the simplest meta-
phors, exactly what the content is supposed to be'' (WMM 262). And
now that we have a more precise characterization of the (purported)
content or truth-conditions of a metaphor, more precise formulations of
the same challenge will hover over us for the remainder of this book. In
chapter 7 I shall return to directly address various ``transcendental''
arguments for this skeptical challenge.33 In this section, I shall focus on
two issues concerning ``mechanics'': (i) constraints that render it at least
plausible that metaphorical character yields determinate contents for at
least some metaphors in their respective contexts; and (ii) the further
question of whether those constraints are semantic. Let me begin by
locating the context-dependence specic to metaphorical character among
three (or four) distinct kinds of context-dependence that bear on meta-
phorical comprehension in general; these correspond to the presemantic,
semantic, and postsemantic stages I introduced in chapter 2.
At the rst stage, we are presented with an event we take to be an in-
stance of a use of a given language; our task is to assign it the linguistic
description of an expression type, including a character. Among alter-
natives, we must determine, or select, whether the utterance is to be
interpreted literally or metaphorically and, in the latter case, which con-
stituents of the utterance should be interpreted metaphorically, that is,
assigned metaphorical charactersthe nonconstant characters of their
underlying metaphorical expressions. We can make this picture of selec-
tion or determination among alternative expression types more explicit in
terms of the structure of a grammar.
Suppose the grammar assigns to every string S under a structural de-
scription a character that, in a context (or in conjunction with other
``cognitive faculties'' or the subject's central belief system), determines the
content of S.34 Suppose also that the lexicon (taken as part of the base of
the grammar) contains a metaphorical-expression-generating ``operator''
(to the status of which we'll return in ch. 6) `Mthat' such that, for every
expression F of constant character (at least with respect to the meta-
phorically relevant contextual parameter), the resulting metaphorical ex-
pression `Mthat[F]' has a nonconstant character.
For every string S to which the grammar assigns at least one character
that contains no character of a metaphorical expression, let it also gen-
erate a set of characters whose members are the characters of all those
strings that result from forming all grammatically admissible metaphori-
134 Chapter 4

cal expressions, and combinations of metaphorical expressions, from the

expressions in S. (For simplicity, I include among these the character of
the string containing no metaphorical expressions; we might think of this
as the literal character of S.) Call this the metaphor set of characters for S.
So, for each metaphorical expression `Mthat[F]' (and combinations of
such expressions) corresponding to an expression F in S, there is a corre-
sponding character in the metaphor set of S. (Here I use double angles
``hh. . .ii'' to represent the composite characters of sentences formed from
the sequentially ordered characters of their constituents.) For example,
corresponding to the string (14) and the character for its ``literal'' inter-
pretation, represented by (14a), the grammar will also generate a meta-
phor set of characters for (14be):
(14) Juliet is the sun.
(14a) hh{Juliet}, {is the sun}ii.
(b) hh{Juliet}, {Mthat[`is the sun']}ii.
(c) hh{Mthat[`Juliet']}, {is the sun}ii.
(d) hh{Mthat[`Juliet']},{Mthat[`is the sun']}ii.
(e) hh{Mthat[`Juliet is the sun']}ii.
Most of the characters in its metaphor set will never actually be assigned
to any utterance of (14) as its (metaphorical) interpretation. But for every
expression (type) in every sentence which it is grammatically possible to
interpret metaphorically, there will be such an array of charactersits
metaphor setcontaining the characters of the corresponding metaphor-
ical expression(-type)s generated by the grammar of the language. Each of
(14ae) is, in other words, the character of an alternative type of which an
utterance of (14) might be identied as a token. This is the sense in which
a metaphorical expression `Mthat[F]' underlies an utterance of an ex-
pression F that can be interpreted metaphorically: To interpret the utter-
ance u of F metaphorically, one must ``recover'' or assign u the type of
the metaphorical expression `Mthat[F]' with its nonconstant character.
Some utterances admit more than one possible structural analysis as a
metaphor, ambiguities that are sometimes exploited by authors and poets
with great eectiveness. The notational use of `Mthat' oers a convenient
linguistic device to represent these ambiguities. For example, it may be
ambiguous whether it is the subject term or predicate in the utterance that
is to be interpreted metaphorically. An utterance of
(15) The owers are smiling in the light
Knowledge of Metaphor 135

is interpretable either as an utterance about the play of sunlight and

shadow on owers, that is, as
(15a) The owers Mthat[`are smiling'] in the light,
or as a comment, perhaps with sexual innuendo, about some pretty young
girls:
(15b) Mthat[`The owers'] are smiling in the light.
Likewise, the scope of the metaphor in an utterance may be (deliberately)
ambiguous, another structural feature that we can mark using `Mthat'.
For example, in a famous passage in Crime and Punishment, Raskolnikov
confesses to Sonia that he has murdered the old money-lender, an act he
attempts to justify by depicting his murder victim (perhaps validly) as an
utterly worthless human being who merely lived o the suering and pain
of others, a parasite of no value whose elimination will even be a benet
to everyone else.35 Here is the subsequent dialogue between Raskolnikov
and Sonia:
(16) R: I've only killed a louse, Sonia, a useless, loathsome, harmful
creature.
S: A human beinga louse!
The exchange plays on an ambiguity in Raskolnikov's utterance to which
Sonia strongly reacts. Raskolnikov wishes to justify his action of murder
by (metaphorically) calling the money-lender `a louse'. Yet, even if the
money-lender was no better than a louse, Raskolnikov would have only
been justied in saying:
(16a) I've only killed Mthat[`a louse'],
an act, as Sonia replies, that would still count as killing someone, that is,
murdering a human being. Raskolnikov, however, seems to believe that
his act would be better (structurally) described in dierent terms, or at
least he attempts to depict it as nothing more than a matter of:
(16b) I've only Mthat[`killed a louse'],
meaning: I have performed an act that should be judged as no worse than
the (minor, justiable) one of killing a louse, rather than as killing a
human being. As Roger White has pointed out, the ambiguity might have
been eliminated had Raskolnikov said either
(17a) I only murdered a louse
understood as
136 Chapter 4

(17b) I only murdered Mthat[`a louse'],

or
(18a) I only stepped on a louse
understood as
(18b) I only Mthat[`stepped on a louse'].
Hence, as White concludes, the ambiguity must have been deliberately
intended by Dostoyevskyprecisely to highlight the moral ambiguity of
Raskolnikov's action. Be that as it may, for our purposes, the important
point is that the structural, linguistic basis for these dierent inter-
pretations of the utterance is clearly marked by the scope of the Mthat
operator. In identifying the utterance as a metaphor, we must assign one
or the other character to the utterance as its typeor, as in this example,
acknowledge both, leaving open the proper description of the action.
This taskthe identication of an utterance as a metaphor by selection
among the alternative characters in its metaphor set, all of which are
made simultaneously available to us by the grammaris what I called
our ``knowledge that metaphor'' in chapter 1. To make the identication,
we employ all sorts of contextual cues, shared background beliefs, and
assumptions about the discourse.36 Here context plays a denite role but,
although it aects what is ultimately expressed, its contribution at this
stage is no part of what is said. In our earlier terminology, this role of the
context is presemantic.
Suppose now that we have identied at least one constituent of the
utterance as an expression to be interpreted metaphorically; that is, we
have assigned the utterance a character from among (14ae). So, at least
one element in the utterance is assigned the linguistic type of a metaphori-
cal expression, that is, the form `Mthat[F]' whose nonconstant character
yields dierent contents given dierent sets of metaphorically relevant
presuppositions. At this second stage, the context furnishes the relevant
parameter to which the character of the metaphorical expression is sensi-
tive, a context set of presuppositions, the value of which is mapped into
the content of the metaphor. If the speaker's knowledge of character is
semantic, this role of the context is also semantic. But it is at this stage
that the force of the skeptic's question is felt; I'll return to it in a moment.
At a third stage, once the content of the metaphor is xed in a particu-
lar context c, the context functions again, this time to evaluate the meta-
phor. Here, however, we should distinguish two kinds of evaluation. The
rst kind of evaluation is whether what is said by the metaphor in the
Knowledge of Metaphor 137

context c is true or false in c. The relevant feature of the context for this
task is the actual circumstance of the context, not (as it was at the second
stage) what is presupposed. But what is ``actually the case'' is not abso-
lute. The evaluation of an interpretation as true or false itself often
depends on contexton the purpose of the utterance and the speaker's
intentions, the expectations and knowledge of its audience. However, this
kind of context-dependence aects the metaphorical no more and no less
than the literal. To use Austin's well-known example, `France is hexago-
nal' is true in many contexts and for many purposes``good enough for a
top-ranking general, perhaps''but false in and for others``not [good
enough] for a geographer.''37 Similarly, `Juliet is the sun' under its meta-
phorical interpretation may be true in some situationssay, Romeo's,
given his expectations and intentionsbut false in other circumstances
say, that for a Shakespearean college admissions ocer.38 This role of the
context must also be distinguished from its semantic role at the second
stage. The argument for the semantic context-dependence of metaphors
turns on their systematic and constrained productivity: the fact that dif-
ferent tokens of one metaphorical expression type, given dierent context
sets of metaphorically relevant presuppositions, express dierent contents.
In contrast, truth-evaluations, as Austin says, are typically ``rough.''
Sometimes we demand strict conformity or correspondence to ``the facts,''
sometimes less stringent standards. But the same context-dependent vari-
ation in standards of truth arises equally for evaluations of literal utter-
ances and metaphors. This role of the context is therefore postsemantic
insofar as the standards of evaluation (including that of truth) will vary
with the further extralinguistic functions of the utterance.39
The second kind of evaluation is of the appropriateness of the assertion
of the metaphor; here too, context plays a central role. This kind of evalu-
ation is broader than the rst because the actual truth-value of the utter-
ance in its contextor what the truth-value is believed to beis only
one among a variety of factors that aect judgments of appropriateness.
A second set of considerations concerns whether or not the content is
informative, interesting, redundant, relevant, novel, tired, and so on. A
third type of consideration is whether the metaphorical character of the
utterance, given the presuppositions held in the context, is an appropriate
medium for the expression of it contentwhether the metaphorical mode
of expression in its respective context is eective, too clear or too obscure,
accessible or not, witty or trite, and so on. These desiderata will obviously
vary with the illucutionary and perlocutionary uses of the utterance and
its audience. To give one example, nowadays we tend to rate (at least
138 Chapter 4

nonpoetic) metaphors higher if they are relatively accessible and more

easily interpretable, but for some ancients and medievals the great virtue
of some metaphors was their power to obscure, conceal, or hide meta-
physical or theological ``secrets.''40 This was also their explanation for
certain metaphors and gures found in Scripture and other sacred writ-
ings. In these texts, the more transparent and penetrable the metaphor,
the worse they judged it to be.
At this third stage, then, the context set of presuppositions bears, not
on the determination of what is saidas it did at the second stagebut
on its evaluation: whether the content is or isn't appropriate. Hence this
role of the context is also postsemantic. However, the story is still more
complicated. As noted in section III, appropriateness (relative to the I-set
of presuppositions) also comes into play, though less directly, already at
the second stage: in determination of content. But there is a dierence
between the two roles of appropriateness. At the third stage we ask
whether what is said is an appropriate thing to say in its context; at the
second stage we ask whether some content can be appropriately expressed
by the metaphorical expression in that context. These latter criteria of
appropriateness ll in, I now want to propose, the constraints on the ap-
plication of metaphorical character: They furnish conditions that restrict
the properties generated as the content for the metaphor in a context.
However, as I said earlier, if these are indeed the constraints on inter-
pretation, there follows a further question: Are they semantic? To see
the problemwhether presuppositional appropriateness can semantically
constrain interpretationlet's look more closely at the ``mechanics'' of
the presuppositions at the second stage.
Back in chapter 1, we observed the divergent interpretations of the
members of each of the groups (repeated here) (1922) and (2324):
(19) Juliet is the sun.
(20) Achilles is the sun.
(21) Before the sun of Moses had set, the sun of Joshua had risen. (BT
Qedushin 72b)
(22) The works of great masters are suns which rise and set around us.
The time will come for every great work that is now in the descendent to
rise again. (Wittgenstein 1980)
(23) Life is a bubble.
(24) The earth is a bubble.
Knowledge of Metaphor 139

At a rst glance, we said, these examples suggest that interpretations of

the same metaphorical expression (`is the sun') vary systematically be-
cause of their dierent linguistic contexts. But it is easy to see that the
relevant context is not (merely) linguistic. Consider Herman Hesse's
(25) He is a shadow over the land,
which, depending on context, might be taken to say either that the subject
is a menace and source of gloom or that he is a shade, protection, even
relief from a devastating threat or power. Here the linguistic context of
(25) remains the same while the interpretation of the metaphor `is a
shadow over the land' varies. So, the metaphorically relevant parameter
cannot be its linguistic context per se; it must be the extralinguistic pre-
suppositions associated with the linguistic items. Furthermore, the ``met-
aphorically relevant presuppositions'' are not limited to those directly
related to, or m-associated with, the literal vehicle of the metaphor (e.g.,
`is the sun'); just as important are the presuppositions related to the rest of
the linguistic frame of the utterance. In fact, as the variations in the in-
terpretation of (25) suggest, the relevant presuppositions might also em-
brace presuppositions manifesteven manifestableonly nonverbally,
hence not directly linked to any verbalized expression in the utterance.
These latter presuppositions are also as heterogeneous in content as those
associated with the metaphorical expression and, whether they are com-
munity-wide (e.g., about stereotypical or normal features) or idiosyn-
cratic, they are not generally part of anyone's linguistic knowledge. Along
with those associated with the literal vehicle of the metaphor, I shall
therefore also add these to the context set of presuppositions, among the
I-presuppositions.
However, there is one signicant dierence between the roles of the
presupposed properties m-associated with the literal vehicle of the meta-
phorical expression and those associated with its environment. Using the
rst set of presuppositions, we generate the (sets of ) properties from
among which the interpretation of the metaphor is constituted. But these
presuppositions typically overgenerate: Some of the (sets of ) properties
presupposed to be m-associated with the literal vehicle of the metaphori-
cal expression are inappropriate as the content of the metaphor in its
context. Using the second set of presuppositionsassociated with the
linguistic and nonlinguistic environment of the metaphorwe lter out
those unsuitables. I will therefore call the rst set, associated with the
metaphorical expression itself, the productive (p-) presupposition set, and
the second set, associated with the environment, the lter (f-) set.41
140 Chapter 4

The way we use the f-presuppositions to constrain metaphorical inter-

pretations is primarily by application of the same two principles intro-
duced in section III to govern the A-presuppositions: Inconsistency and
Redundancy. Suppose (to anticipate an idea I shall discuss at length in the
next chapter) that the properties that `is the sun' in (19) uttered in c met-
aphorically expresses are those that are presupposed to be exemplied
or sampledby the sun in the context c.42 These properties might in-
clude, to begin with, properties such as being the major source of direct,
natural light (in contrast to the moon, stars, or a lamp). This property is
inconsistent with what we know of human beings and therefore would be
inappropriate to assert of Julietand, for the same reason, will be
excluded from the interpretations of (20) and (21) and, for a similar rea-
son, from (22). But more specic to (19), and possibly as opposed to (20)
and (21), among the properties exemplied by the sun in c might be the
property of being the one that rises and sets faithfully and regularly, day
after day without fail, as well as being the one worthy of worship. Both
properties are equally generated p-presuppositions but only the latter,
not the former, would be an appropriate property to assert of Juliet
in Shakespeare's context. For Romeo's Juliet never declines; she is
f-presupposed not to be like other natural, ordinary creatures who rise
and wane with time and nature, whose day inevitably ends, like the setting
of the sun, after it has run its natural course. (Hence the tragedy of her
death, perhaps, and the irony of Romeo's metaphor.) Therefore, to assert
that p-presupposed property associated with the metaphorical expression
would contradict our other f-presuppositions associated with its environ-
ment (i.e., associated with `Juliet'), violating the rule of Inconsistency not
to interpret an expression in such a way that what would then be asserted
is already presupposed to be false. But the context might have been
otherwise. In the Talmudic context of (21), it is presupposed that Moses
and Joshua, for all their illustriousness, are equals of a kind; surely that
neither ought to be worshipped. Given that presupposition and applica-
tion of Inconsistency and perhaps Redundancy (which can also function
to promote a property otherwise not heavily weighted), the metaphor `is
the sun' in (21) is awarded as its content something like the property of
proceeding in uninterrupted succession and continuityrise, followed by
fall, followed by rise.
It should be noted, rst, that in all these examples the f-presuppositions
interact with the p-presuppositions through the medium of general prin-
ciples of assertion; there is no need to introduce any special techniques of
interpretation specic to metaphor.43 Second, and to avoid a potential
Knowledge of Metaphor 141

misunderstanding, I have described the relation between the p- and f-

presuppositions as if the second temporally follows the rst, but they need
not interact in a xed order. In impressionistic terms that I'll make more
precise in chapter 5, the f-presuppositions not only discourage certain
interpretations already generated by p-presuppositions; they may block
the generation of some and indirectly encourage the generation of others.
For example, by limiting the appropriate sets of m-associated features to
those attributable to animate and human subjects (presuppositions trig-
gered by our knowledge of the proper names that occupy the subject
positions in (19)(22)), the f-presuppositions already preselect the range of
features generated by the corresponding p-presuppositions. As it were,
certain features are so far aeld that they seem never to arise as potential
interpretations; others are eliminated as inappropriate although they are
features ``in principle'' sampled by a scheme. In the former case, we might
think of the f-presuppositions as operating before the p-presuppositions.
In sum, conditions of appropriateness embodied in rules like Inconsis-
tency and Redundancy constrain applications of the rule of metaphorical
character already at the semantic stage. Nonetheless, the skeptic may still
not be convinced that these constraints suce to generate determinate
contents for arbitrary metaphors. Even where they do, he may also object
that Inconsistency and Redundancy are general principles of rational
action, not semantics-specic. So, either the speaker's exclusively seman-
tic knowledge of character does not yield knowledge of determinate con-
tents or, to make the contents determinate, we must supplement character
by extrasemantic conditions.
Two comments in reply: First, at the third postsemantic stage the po-
tentially pertinent presuppositions can include virtually any presupposi-
tion one might introduce, not just the relatively local productive and lter
presuppositions of the second stage, presuppositions associated with par-
ticular constituents of the utterance in question. Therefore, the pre-
suppositions of the last stage are global and Quineanno belief can be
absolutely ruled out as nongermanewhereas those that are active at the
second stage are (relatively) locally partitioned to the task at hand.44
Second, rules like Inconsistency and Redundancy do not belong to the
constitutive conditions of metaphorical character. They are also not
semantics-specic. But they also dier from the various criteria of appro-
priateness to which we appeal at the third, last stage that depend on the
illocutionary and perlocutionary uses of the statement. As general con-
ditions on rational behavior that apply to all applications of the rule of
character, Inconsistency and Redundancy are rules that all speakers in all
142 Chapter 4

linguistic or speech communities follow and to which their linguistic be-

havior should conform. Although they are not rules of language, they are
rules that any speaker of a language must know.
Neither of these comments is a knock-down argument for the thesis
that it is the semantic character of the metaphoras opposed to or au-
tonomous of all pragmatic considerationsthat yields a unique, xed,
determinate set of properties as its content or (truth-conditional) inter-
pretation. I have also described the dierent stages as if they were linearly
ordered in real time; in fact, the various kinds of context-dependence
continually interact. Apart from the interaction of the dierent kinds of
appropriateness at the third stage, considerations of appropriateness of
the last kindwhether a metaphorical character is an appropriate way to
express a given contentalso bear on metaphorical identication at the
rst stage.
Yet another issue complicates the question whether metaphors have
determinate, semantically xed contents. The propositional content of a
metaphorthe theoretical explication of what is said by the utteranceis
not always distinguished, or distinguishable, from what is asserted and
what is intended by the utterance. And in those situations where it makes
a dierence whether a metaphor has a determinate contentwhere we
are called upon to adjudicate disagreementsit is not always clear
whether the disagreement is over what is said, what is asserted, or what is
intended.
The dierence between what is said or asserted and what, in addition, is
also intended is clear. Let me illustrate the ambiguity I have in mind be-
tween what is said, or content, and what is asserted. What an utterance
asserts is dened for us in contrast to what it presupposes. But depending
on the trade-o between assertion and presupposition, what a given
utterance asserts may or may not turn out to be identical with, or individ-
uated along the same lines as, its propositional content. For example, I
assume that the content, or truth-conditions, of the predicate `is a bache-
lor' is the complex property of being an unmarried male.45 Suppose now
that I ask you whether the man standing to your right is married and you
answer, ``No, he is a bachelor.'' Your utterance presupposes that the
subject of the sentence is a male and it asserts that he is unmarried. In this
case, what is asserted is part of but not identical with what I assume (even
in this case) is the truth-condition, or content, of the predicate.
A similar ambiguity aicts our talk about what is said/asserted by a
metaphor relative to its context set of presuppositions, especially its f-
presuppositions. What is asserted by a metaphor in a context is a function
Knowledge of Metaphor 143

of the postsemantic considerations of appropriateness (at the third stage)

as much as it is of the kinds of appropriateness that bear on determination
of semantic content (at the second stage). Depending on how sharply we
wish to distinguish what the metaphor asserts from what is presupposed
in the context, we can draw the circle more or less tightly around its
content. So long as it is ambiguous whether we are concerned with
what is asserted or with what is said (i.e., content) by the metaphor, it
will be correspondingly unclear whether its ``interpretation'' is or is not
determinate.
I hope I have said enough to allay a skeptic's doubts that the character
of a metaphor never yields a determinate content in its context and that,
whether we call them semantic in a strict sense or not, the constraints that
generate metaphorical contents are suciently dierent from the clearly
nonsemantic ones we use to evaluate their postsemantic appropriateness.
In the next chapter, I shall turn to more pragmatics to ll out our se-
mantic story and illustrate how the semantics interacts with other aspects
of the context in metaphorical interpretation. In chapter 6 I'll return to
implications of our semantic knowledge of metaphor, both in order to
work out remaining details and to illustrate its explanatory power in
articulating various distinctions and notions that run through our pre-
theoretical concept of a metaphor.
Chapter 5
Knowledge by Metaphorical
Content

In chapter 4 I laid out the basic theory of our semantic knowledge of

metaphor and the structure of the presuppositions that constitute the
context of a metaphor. I located the pertinent notion of presupposition
among the attitudes and explained the sense in which presuppositions
are presuppositions for an interpretation; the degree to which its presup-
positions are required for the interpretation of a metaphor; and the vari-
ous roles of presuppositions in determining an appropriate metaphorical
interpretation of an expression. In chapter 6 I'll return to metaphorical
character and its status as meaning. In this chapter I'll discuss in greater
detail how we apply our semantic knowledge to particular contexts, that
is, particular kinds of context sets of presuppositions, to generate content.
In the terminology of chapter 1, this content is one type of knowledge (we
convey) by metaphor. A second type, to which I'll return in chapter 7, is
knowledge (conveyed) by metaphorical character.
Insofar as these issues are a matter of how speakers apply their seman-
tic knowledge to particular extralinguistic contexts, or use their semantic
competence specic to metaphor, they fall in the category of pragmatics
of metaphorical interpretation. In more traditional terminology, the issues
concern what are called the ``grounds'' of metaphorical interpretation, for
example, the role of similarity judgments. In terms of my theory, these
grounds, which are typically not semantics- or language-specic (and in-
troduce additional dimensions of context-dependence), furnish the con-
tent of metaphorically relevant contextual presuppositions.
Without this pragmatic story we have no explanation of how an actual
utterance of a metaphor has an interpretationpropositional contentin
a particular context. Working through these issues will therefore enable us
to see how the theory is put into practice. But the point of this chapter is
not to test the adequacy of our semantics or to provide recipes to generate
interpretations of metaphors. The semantics supplies only the form of, or
constraints on, metaphorical interpretations; it does not predictand
146 Chapter 5

therefore does not stand onthe actual interpretation of any metaphor.

That will ultimately depend on the ingredients we put into the context set
of presuppositions.
My aim in the chapter is, by showing how our semantic competence in
metaphor interacts with our extralinguistic contextual presuppositions, to
highlight certain dierences between our knowledge of character-rules
and the kinds of abilities or skills involved in their use or application
to contexts. The semantic character-rules apply to all metaphors, while
dierent metaphors rest on dierent grounds that yield dierent m-
associated features. The diversity of these latter, pragmatic features has
persuaded some theorists that there is no general rule-governed explana-
tion of metaphor, only case by case stories about individual metaphors.
But this is because they focus only on the pragmatic aspects of meta-
phorical interpretation. This is also not the end of their story. When they
next add the assumption that literal meaning, in contrast, is a function of
linguistic competence, they conclude that there is an untraversable dis-
tance between the literal and metaphorical. The moral I want to suggest
instead is that the distance is not between the metaphorical and the literal
but between the semantic knowledge underlying metaphorical interpreta-
tion and the nonsemantic skills that ground their presuppositions.
I have already indicated (ch. 4, sec. I) how wide and varied is the range
of m-associated features, corresponding to dierent grounds, that under-
lies the interpretations of dierent metaphors. I shall not review this here.
I shall concentrate on one class of metaphors based on exemplication, a
symbolic relation that I'll argue is a source of the (or one class of ) simi-
larity judgments that are said to ground metaphors. Despite a venerated
tradition, similarity has taken a beating during the last thirty years, not
only in connection with metaphor but with many topics such as pictorial
representation. Yet its critique has signicantly improved our under-
standing of the relation. I shall briey trace the history of this discussion,
which will lead us to exemplication. My analysis of exemplication, in
turn, will introduce the role of networks and systematic family-sized units
in metaphorical interpretation, among which I'll distinguish three kinds.
This notion of a family of metaphors will, nally, play a crucial role in my
explication of knowledge by metaphorical character in chapter 7.

I A History of Similarity in Metaphor

Historically, what I shall call the the similarity thesis:

(i) Similarity and comparison judgments play a signicant, if not
essential, role in metaphorical interpretation
Knowledge by Metaphorical Content 147

has been associated with a second thesis:

(ii) Metaphors are elliptic similes.
I shall return to discuss (ii) in chapter 6 but, in any case, it is important to
distinguish (i) from yet a third thesis, also associated with (ii):
(iii) Metaphors (i.e., utterances of declarative sentences containing at
least one metaphor) assert similarities.
The similarity thesis (i) contains an important grain of truth even if it is
not the whole truth, if only because the m-associated features that enter
into metaphorical interpretations include many that do not arise from
similarity judgments.1 But, contra (iii), even those metaphors grounded in
similarity judgments need not, and do not, assert similarities. The content
of the metaphor `Juliet is the sun' is not that of a relational statement such
as `Juliet is like the sun' or `Juliet resembles (is similar to) the sun'.2 If it
were, the logical form of the metaphor whose surface grammatical struc-
ture is Sj would be that of a two-place relation Rjs. But without compel-
ling evidence to the contrary, there is no reason to abandon the surface
grammatical form of the metaphor, which is that of a one-place predica-
tion in which we say something about something or attribute the property
expressed by the predicate to the subject-referent.
Where, then, is the similarity judgment located? In the context of the
metaphor rather than as part of its content: The similarity judgment gen-
erates the presuppositions about the relevant features or properties m-
associated with the literal vehicle of the metaphor. The corresponding
constituent of the content, on the other hand, is simply the attributed
feature or property, not the similarity relation that grounds the context set
of presuppositions to which the character of the metaphorical expression
`Mthat[F]' is sensitive.3 Throughout this chapter, I shall therefore con-
tinue to assume that ``similarity theories'' of metaphor are theories about
similarity only as a source of contextual presuppositions, setting aside the
logical form of the metaphor, which I'll assume is that of a predication
rather than a relational statement.4
With this role of similarity in mind, any theory of metaphor that
employs the notion must address three standard objections.5
1. Any two objects share some common property by which they could
always be said to resemble each other; therefore, to avoid the charge of
vacuity, such theories owe us additional criteria to explain which proper-
ties count.
2. If a similarity judgment is made by comparing independently known
and identiable features of the objects in question, a similarity theory of
148 Chapter 5

metaphor will be able to take into account only those properties known or
identied independently of or prior to the context of the metaphor. It
cannot allow for similarities or properties that the utterance of the meta-
phor, as Black put it (see ch. 4, sec. II), ``creates''even in the epistemic
sense of bringing them to our attention or making us notice themby
likening the referents of its terms.
3. On the widely shared assumption that similarity is a symmetrical rela-
tion, similarity statements should be reversible salva veritate and even
salva signicatione: A is like/similar to B if and only if B is like/similar to
A. Metaphors, however, are not reversible. `This man is a lion' may be
true (under its metaphorical interpretation), but it does not follow that
`This lion is a man' is also true (under its metaphorical interpretation).6
Furthermore, even where the original and its converse both happen to be
true, they generally have dierent meanings: Contrast `My butcher is a
surgeon' with `My surgeon is a butcher'. Indeed the converse is not always
even interpretable metaphorically: Contrast `John's face is a beet' with `A
beet is John's face'. Therefore, the objection concludes, similarity state-
ments and metaphors must have completely dierent logical structures,
excluding one as grounds for the other.7
In reply, let me begin with the third objection, which presupposes a
``geometric'' model of similarity that represents objects as points in a co-
ordinate space whose (dis-)similarity is a function of their metric distance
from each other. (It is arguable that the rst objection also assumes such
a model.) This is not the only conception of similarity in the literature. In
particular, Amos Tversky (1977) proposes a ``contrasting feature match-
ing'' model, according to which the degree of similarity between two
objects is a weighted function of their common and distinct features. An
immediate consequence of this model is that, contrary to the metric
model, some similarity judgments and statements are asymmetric. Thus
Tversky observes that the choice of subject (x) and referent or (as I'll say)
predicate ( y) in ordinary statements of the form `x is like y' or `x resem-
bles y' is not in general arbitrary. We say that the portrait resembles
(is/looks like) a person, but not that the person resembles (is/looks like)
the portrait; that a son resembles (is/looks like) his father but not that the
father resembles (is/looks like) his son; that the ellipse is like a circle but
not that the circle is like an ellipse. Likewise, the judged similarity of Bill
Clinton to JFK is much greater than the judged similarity of JFK to Bill
Clinton. And with similes (or nonliteral similarity statements), the asym-
metry or directionality is even more pronounced. Thus, the metaphor
`John's face is (like) a beet' is interpretable, although `A beet is (like)
Knowledge by Metaphorical Content 149

John's face' is not; and, even where the two are interpretable, as in `A
man is (like) a tree' (i.e., having roots) and `A tree is (like) a man' (i.e.,
having a life history), they have dierent interpretations. In all these
examples, the similarity judgment/statementliteral or notis direc-
tional and asymmetric.8 Thus the central assumption underlying the third
objection fails to hold.
On Tversky's account, a similarity judgment is instead a function of
three factors: our ``feature representations'' of the objects under compar-
ison, the ``focus'' of the similarity judgment, and the salience of the fea-
tures. First, we represent each object a and b by a set of features or
attributes A and B, respectively. Of course, these sets of features by which
we typically represent objects are huge, and only a small number are rel-
evant to a given cognitive task on an occasion. Hence each of the sets of
features that enters into a specic similarity judgment must itself be the
``product of a prior process of extraction and compilation'' (Tversky
1977, 330). Tversky says very little about this process, but I'll return to
this part of the story.
Given feature representations for each object, Tversky argues that the
perceived similarity of two objects a and b [s(a; b)] is a weighted function
of their shared salient features less the sum of a weighted function of the
attributes distinctive only of one and a weighted function of the attributes
distinctive only of the other.9 What determines the salience of a feature?
Tversky initially suggests a wide variety of sources``intensity, fre-
quency, familiarity, good form [for geometric gures], and informational
content'' (ibid., 332)implying that there are no general principles that
govern such judgments. However, later in the essay Tversky proposes two
main determinants of salience; the second of these is context-dependent,
the rst not:
1. Strengththe intensity of the object or its signal-to-noise ratio (e.g.,
brightness of a light, loudness of a sound, saturation of a color)which is
relatively stable across contexts and independent of the object of which
the feature is an attribute.
2. Diagnostic Value: the classicatory signicance or importance of the
feature. We sort objects so as to maximize the similarity of those in a class
and their dissimilarity from objects outside the class. Features that yield
better classications of this kind have a higher diagnostic value and hence
higher salience. But the discriminability of an object by a feature is highly
sensitive to contextthe function of the classication, the identity of the
object, and the realm of objects to be sorted on the occasion. Hence this
determinant of salience is context-dependent.
150 Chapter 5

Finally, Tversky argues that we assign dierent weights to the distinc-

tive and shared features of the things being compared, depending on the
``focus'' of the similarity judgment. This relative weighting explains why
some similarity judgments are symmetric and others not. When the judg-
ment assesses the degree to which a and b are similar to each other,
focusing no more on a than on b, the distinctive and shared features of
both a and b are equally weighted and the judgment is symmetrical.
However, when we judge the degree to which a is similar to b, Tversky
hypothesizes that we focus on the subject a, weighting its distinctive fea-
tures more than those of the predicate b; hence the distinguishing features
of the subject reduce the degree of similarity more than those of the
predicate. But we also tend to choose as predicate a term whose referent
has the more salient features, which thereby increases the degree of simi-
larity, and as subject the term whose referent is the less salient object. For
example, the variant in a category is said to be similar to the prototype of
the category, but not vice versa. Thus the asymmetry of many similarity
judgments is due to the fact that when we judge that a is similar to b, the
weighted salient features of b shared by a must be greater than the
weighted distinctive features of a and of b. However, when we reverse this
order of subject and predicate, and judge whether b is similar to a, we
change the focus of the judgment and hence the weighting.
As mentioned earlier, Tversky's observation that many similarity
statements are directional and asymmetrical counters the main assump-
tion of the ``reversibility objection'' to the comparison theory of meta-
phor.10 But Tversky also makes an important positive suggestion about
the relation of similarity to metaphor. With similarity judgments, he says,
we assume a particular feature space and ``assess the quality of the match
between subject and referent''; with interpretations of metaphors and
similes, we assume a resemblance ``and search for an interpretation of the
space that would maximize the quality of the match'' (ibid., 349). This
proposal suggests partial replies to the other two objections with which
we began. If the quality of a match is a function of the salience of the
featureswhich in turn are a function of their intensity and diagnostic
valuewe now have criteria to determine which features, relative to the
context, count toward determining the relevant similarity. And insofar as
the ``interpretation of the feature space'' is subsequent to the assumption
that there is similarity at work, there is the germ of an account for Black's
creative metaphors.
This last proposal aside, it should be emphasized that Tversky's pri-
mary concern is similarity, not metaphor. For our purposes, his most
Knowledge by Metaphorical Content 151

important contribution is his substantive account of the criteria of

salience, the rst such account to be proposed. However, other psycholo-
gists and cognitive scientists have attempted to modify Tversky's account
in order to directly address metaphors and nonliteral similarity statements
like similes. To do this job, perhaps the most serious problem facing
Tversky's theory is its apparent inability to explain why metaphorical
comparisons are not only asymmetric and directional (like all comparison
statements) but truly irreversible: Contrast the reversible though direc-
tional literal comparison `Taiwan is like Japan' with the nonliteral `Israel
is like David ghting Goliath' whose reverse is not even interpretable. To
account for this dierence, Andrew Ortony (1979) next proposed that the
salience of a feature is a function not only of its intensity and diagnostic
value but also of its ``importance,'' or stereotypical value, for the object of
which it is an attribute. For example, redness is a much more ``important''
or stereotypically characteristic attribute for our notion of retrucks than
for bricks, even though it is frequently true of both and it is necessary for
neither. However, this criterion entails that, even apart from diagnostic
purposes, the salience of a feature is sensitive to the feature representation
of the object of which it is an attribute; hence the salience of one and the
same feature may dier in the feature representations of dierent objects.
And this, in turn, raises the question how to compute the salience of a set
of shared features in a representation. In particular, where there is signif-
icant ``salience imbalance'' between the feature's salience in A and B, is
the salience of the intersection an average of the two or just that of either
A or B? Ortony proposes that it is a function of the salience value of the
feature in B alone.11 It follows that if the feature has high salience in B
but not in A, the measure of the degree of similarity of A to B will be
signicantly higher on Ortony's account than on Tversky's.
This fact, Ortony argues, explains why nonliteral similarity statements
such as `billboards are like warts' are not just asymmetric but irreversible.
The feature ``being ugly'' is stereotypically signicant for and hence a
highly salient feature of warts but not of billboards. Hence there is con-
siderable salience-imbalancelow salience in A, high salience in B. Its
high salience in B in turn increases the weighting of the shared properties,
including ugliness, of billboards and warts, and thus their degree of simi-
larity. In contrast, in `warts are like billboards' there are no comparable
shared properties with high salience in the representation of `billboards'
and low salience in the representation for `warts', decreasing the degree of
similarity. Therefore, it is impossible to reverse the original nonliteral
statement while preserving the same high measure of similarity. On the
152 Chapter 5

other hand, with a literal similarity statement like `billboards are like
placards' in which the high-salience features of placards (e.g., being a
printed poster, for display in public places) are also high-salience fea-
tures of billboards, there is relatively little salience imbalance. Hence the
statement is reversible despite some dierence in degree of salience and
perhaps a change of meaning. Generalizing from these cases, Ortony
hypothesizes that the greater the ``salience imbalance''that is, the
greater the dierential in relative increasing salience of a feature from the
A to B representationsthe more metaphorical, or nonliteral, and the
more irreversible we judge the statement.12
One moral of this story is that literality and metaphoricity are not
discrete values but two poles on a continuum of degrees of salience-
imbalance. A second, more important lesson for our purposes is that the
key notion in this analysis of metaphoricity is salience (and salience-
imbalance), not comparison or similarity. Although Tversky rst brought
salience onstage to explain certain features of similarity statements and
judgments, and Ortony continues to maintain that metaphors should be
``analyzed'' as ``indirect'' comparison statements, when we actually turn
to their explanations of metaphoricity all the work is done by the notions
of salience and salience imbalance. The next step in the argument is just
to drop the detour through similarity statements and explain metaphors
directly in terms of salience or a related notion. This, as I understand
their position, is the spirit of a third account recently proposed by Sam
Glucksberg and Boaz Keysar (1990) (henceforth G&K).13
G&K raise two objections against Tversky and Ortony. First, they
point to a disanalogy between literal and metaphorical (e.g., simile) simi-
larity statements. Those who argue that metaphors of the form `a is an F'
are (elliptic) similarity statements frequently point to the possibility of
``inserting'' `like' to yield a (literal or metaphorical) similarity statement.
G&K point to an important dierence in the other direction. In the literal
case, we can never infer `A is B' from `A is like B': `apples are like pears' is
true, but `apples are pears' is false. With metaphorical similarity state-
ments, however, we can always infer `A is B' from `A is like B'; for ex-
ample, `Juliet is the sun' from `Juliet is like the sun' or `sermons are
sleeping pills' from `sermons are like sleeping pills'.14 This disanalogy
suggests, as I'll argue in chapter 6, section VII, that Aristotle's famous
statement that ``the simile is also a metaphor'' should be understood ex-
actly as it is written: Both metaphors and similes are ``metaphorical''
unlike literal similarity statements. Furthermore, it follows that what
renders similes and metaphors metaphorical must be something that
Knowledge by Metaphorical Content 153

enters into a similarity judgment, such as salience, and not something that
should be identied with similarity.
Second, I mentioned earlier that not all of the innumerable, many of
them irrelevant, features of the subject and referent are subject to the
matching test in making a similarity judgment. For this reason, you will
recall, Tversky posits a prior process to select and extract the relevant
features to be put to the matching test. To this G&K object that, if
salience plays a role in the matching task, then itor whatever analogous
quality determines which features ``come readily to mind'' (G&K, 4)
might equally well play a role in the earlier stage of extraction of the rel-
evant features. (To distinguish the weights of the subject and predicate,
G&K, like Tversky, also assume a ``focusing hypothesis'' or similar prin-
ciple based, say, on the given/new or topic/comment distinction.) Thus we
can explain the asymmetry, and even irreversibility, of metaphors and
similes in terms of the role of salience in dierential feature selection
rather than feature matching, thereby entirely eliminating the extra step
through comparison statements. In sum, on all these theories the impor-
tant lesson is that it is salience rather than similarity that emerges as the
key notion for the analysis of metaphor.

II Exemplication

I have now mentioned three determinants of the salience of a feature: (i)

its context-independent intensity; (ii) its context-sensitive diagnostic or
classicatory value; and (iii) its importance within our stereotypical de-
scription or intuitive theory of a thing, its place in our normal notion of
the referent, the socially shared way of thinking of the thing that is typi-
cally acquired with the linguistic expression for that thing. Of these three
sources of salience, the second and third clearly enter into the inter-
pretations of metaphors although there are no hard and fast rules that
associate one source of salience with a particular kind of metaphor. There
is, however, one dierence between the two. Normal notions are typically
associated with property and relational terms, natural kind terms, and
names of public gures; and within a community, they do not vary
signicantly from speaker to speaker or utterance to utterance. Where an
expression has a normal notion associated with it, its features are there-
fore available, at least by default, to serve as a metaphorical interpreta-
tion for the expression. However, because this source of salience is
relatively xed and contextually stable, it will never be a source of pro-
ductive, novel metaphorical interpretations that exploit and vary with the
154 Chapter 5

resources of their individual contexts of utterance. The other, more con-

text-sensitive source of salience, the classicational or diagnostic value of
the feature, is therefore a potentially richer source for metaphorical
interpretations. To spell out its contextual parameters, I will now turn to
exemplication, a mode of reference to features that is one mechanism by
which they are made salient.
The notion of exemplication was rst discussed by C. S. Peirce, but in
recent years the one who has explored it the most deeply is Nelson
Goodman.15 Allowing ourselves a bit more ontological indulgence than
Goodman the nominalist, let's say that an object o exemplies a property
P if and only if (i) o possesses P and (ii) o ``symbolizes, stands for, or
refers to'' P.16 As Pierce already noted and Goodman makes explicit, the
crux of exemplicationand what distinguishes it from mere instantia-
tionis the second condition. It is a mode of reference rather than mere
possession.
Goodman's paradigm of exemplication is the sample, for example, a
tailor's booklet of small swatches of cloth. Each swatch in the collection
possesses an innumerable number of properties but it is a sample only of
some: its color, weave, texture, and pattern, but not its size, shape, abso-
lute value, date of production, or weight. What is necessary for a property
to be sampled is not only that the swatch possess it but also that the
swatch refer to it. I shall not analyze this reference relation. In Good-
man's theory of symbols it is a primitive notion, the most general
mapping from the items of a symbol systemboth linguistic and non-
linguistic, notational and nonnotationalto their correlated eld of
compliants or referents. There is no one explanation for this reference re-
lation, say, in terms of satisfaction of descriptive conditions or in terms of
historical chains. ``An element may come to serve as a symbol for an ele-
ment related to it in almost any way'' (Goodman 1976, 65). To the extent
to which we can nd any account of reference in Goodman's own writ-
ings, it is not semantic but psychological, closer to Morris's conception
than Frege's: ``[E]stablishment of the referential relationship is a matter of
singling out certain properties for attention, of selecting associations with
certain other objects'' (Goodman 1972a, 66). This is, of course, not to say
much more than that the sample makes certain properties salient. So, by
introducing exemplication into the account of salience, my point is not
to reduce the latter to something more basic but rather to distinguish one
of its brands.17
The psychological, or attitudinal, character of this notion of refer-
enceespecially as Goodman describes it in the quotation in the previous
Knowledge by Metaphorical Content 155

paragraphmay help explain two curious facts about the relation. First,
if exemplication is a matter of bringing agents to selectively attend to
something, then, like other objects of the attitudes, what is exemplied
must be individuated more nely than (simple) extensions (objects or
classes); hence it is properties (or, for Goodman, predicates or labels) that
are exemplied. Second, although it is objects that do the exemplifying,
how those objects are described or displayed is also not irrelevant to their
mode of sampling. For example, a creature with a kidney, but not a
creature with a heart, exemplies the type of living organism that elimi-
nates its own waste despite the fact that anything is a renate if and only if
it is a chordate. In this case, however, the description bears not on the
individuation of the exemplifying object but identication of the schema
to which the exemplifying object belongs on the occasion. To explain this
aspect of exemplication, let me turn to its contextual parameters.
Consider again the tailor's booklet of cloth swatches. What determines,
of a swatch's innumerable possessed properties P, those that it also refers
to and those it does not? Each swatch si is a sample not in isolation, but
only as part of the whole booklet S. Two relations between the elements
of S and the eld of properties P prima facie bear on what is sampled by
each si . First, each si in S samples just those features it possesses but the
other sj do not, and, second, each si in S samples just those properties a
contrary of which is sampled by some sj of S. In other words, the features
exemplied by any given object o will depend both on the set of things
presupposed in the context to be the schema S of referring objects to
which o belongs in that context and on the realm of alternative properties
P presupposed to be sorted or classied by the schema as a whole. Sup-
pose the schema consists only of a heavy woolen swatch and a light cotton
one; relative to the alternative features being-a-summer-dress, being-a-
fall-dress, and being-a-winter-dress, the woolen swatch may sample the
two latter and the cotton swatch the rst. But suppose we add a medium-
weight swatch to the schema; the woolen one may now sample only be-
ing-a-winter-dress and not being-a-fall-dress. Likewise, a swatch may be a
sample of red relative to the alternatives red and purple, but be a sample
of pink relative to a feature-space containing red, pink, and purple.
Sampling is a simple case of exemplication, and we cannot generalize
its conditions in any precise way. But it indicates the kinds of desiderata
that characterize exemplication more generally. The example of sam-
pling may also be misleading insofar as it suggests that what something
exemplies always ought to be immediately evident like the features of the
sample that we just ``see.'' This ``visual'' dimension of exemplication is
156 Chapter 5

important: The sample displays what it exemplies.18 But display does

not require instantaneous recognition. For many works of art, it is only
after sustained viewing, with the discrimination of the trained eye, that
the observer can identify what is exemplied. Likewise, it may only be
after considerable analysis and exploration that we can identify what is
exemplied in the interpretation of metaphors (especially) in literature or
poetry.
Finally, it should be mentioned that one object may simultaneously
belong to more than one schema, exemplifying more than one feature
relative to the dierent schemata, all in the same context. The eect can
be dynamic, unpredictable, surprising. Wayne Booth describes how he
once read the concluding paragraphs of Norman Mailer's The Armies of
the Night, two paragraphs full of ``deliberately scrambled metaphors,'' to
a conference on metaphor, intending to elicit the audience's reaction to
the ethical import of the metaphors. To his surprise, the audience instead
laughed. The passage was ``wrenched out of its context, taken out of its
historical moment [the Vietnam war protests] and put into a new context,
where attention was focused on deliberately scrambled metaphor as a
thing in itself '' (Booth 1978, 328). Goodman (1976, 54) gives an example
of a tailor's booklet of swatches that, used in the context of a tailors'
convention, serves as a sample of being-a-tailor's-booklet. Analogously,
Mailer's text is taken here to exemplify the property of being a cluster of
``deliberately scrambled metaphors'' rather than, or in addition to, a text
expressing (among other things) the properties presupposed to be exem-
plied by its multiple metaphors.

III Metaphors of Exemplication

When Romeo utters

(1) Juliet is the sun
I have argued, rst, that its content is not that Juliet and the sun resemble
each other (in some or another respect) and, second, that it is indeed
unnecessary to detour through an underlying judgment of similarity to
determine its content. Instead Romeo is saying that Juliet has a certain
propertycall it Pm-associated with the predicate `is the sun' and that
the crucial explanatory factor that generates P is its salience in the con-
text.19 I now want to propose that, for at least one class of metaphors
(like (1)), the salience of P derives from the fact that it is presupposed to be
exemplied by the object(s) that is (are) the extension of the literal vehicle
Knowledge by Metaphorical Content 157

of the metaphor. Here the sense in which the metaphorical depends on the
literal is that the content of the metaphorical expression `Mthat[`is the
sun']' in a context c is (modulo the lter-presuppositions of the utterance)
the set of properties P presupposed to be exemplied by the extension of
its literal vehicle.
Which properties are exemplied by the referent of the literal vehicle of
the metaphor? What any object exemplies, we said, is relative to the
other elements in the exemplication-schema to which it belongs and to
the eld of features sorted by the schema as a whole. Metaphors of ex-
emplication are no dierent; we therefore need to identify the schema
and eld of features of (1). Let's look again at the context of (1) in
Shakespeare's play, Romeo's speech:
But soft, what light through yonder window breaks?
It is the East, and Juliet Mthat[`is the sun'].
Mthat[`Arise'] fair Mthat[`sun'] and Mthat[`kill'] the envious Mthat[`moon'],
Who is already sick and pale with grief,
That thou her maid art far more fair than she.
Be not her maid since she is envious,
Her vestal livery is but sick and green,
And none but fools do wear it; cast it o.
It is my lady, o it is my love,
O that she knew she were.
She speaks, yet she says nothing; what of that?
Her eye Mthat[`discourses'], I will not Mthat[`answer'] it.
I am too bold, 'tis not to me she speaks.
Two of the fairest stars in all the heaven,
Having some business, do entreat her eyes.
To twinkle in their spheres till they return.
What if her eyes were there, they in her head?
The brightness of her cheek would Mthat[`shame'] those stars,
As daylight doth a lamp; her eyes in heaven
Would through the airy region stream so bright,
That birds would sing, and think it were not night.
See how she leans her cheek upon her hand.
O that I were a glove upon that hand,
That I might touch that cheek.
(Romeo and Juliet II, ii, 223; my italics)
In this passage I have substituted metaphorical expressions for some of
the expressions interpreted metaphorically; I'll return to these later in
this section.20 I have also italicized the terms for a number of light-
related bodies and states that Shakespeare sets up in deliberate contrast
with one another: `the sun', `moon', `stars in all the heaven', `the bright-
ness of a cheek', `lamp', `daylight', and the darkness of `night'. The con-
158 Chapter 5

trasts among the referents of these latter terms are subtle and intricate; I
can touch on only a few of their more obvious connections. The sun and
moon are the two referents placed in the sharpest opposition, perhaps
because `the sun' and `the moon' are the two primary metaphors in the
passage, the one referring, of course, to Juliet, the other to Rosaline,
Romeo's previous beloved whom he abandons for Juliet. But the opposi-
tion between the sun and moon (referred to by the literal vehicles for the
two respective metaphors) rests on a variety of literary and mythological
as well as natural dierences between them. The passage exploits the
mythological, symbolic identication of the moon with Diana, the god-
dess of virginity, who was earlier identied with Rosaline (I, i, 207209.)
and who now is presented as the mistress of Juliet, hence her superior.
Romeo's address to Juliet as the sun recalls earlier lines in which the sun is
said to be worshipped (I, i, 116f.), and is given quasi-divine featuresfor
example, `all-seeing' (I, ii, 94), that is, all-knowing. Alternatively, the sun
may be all-seeing at this point in that all do see it, it is the center of
everyone's attention and eld of vision, and it has nothing to hide or of
which to be ashamedin contrast to the end of the play when ``The sun
for sorrow will not show his head'' (V, iii, 305). In this last line indeed
even the gender of the sun has changed. The usual relation by which the
moon reects, and is subordinate to, the sun is also reversed in our pas-
sage, the sun now being the maid of the moon. But the moon, metaphor-
ically referring to Rosaline, is also described as less ``fair than'' Juliet, the
sun, and envious of herusing the standard of comparison `fair' which
recurs again and again throughout the play (see, e.g., I, i, 204; II, Prol., 3
4)and earlier was used to describe Rosaline (I, i, 219). Romeo calls
upon the sun to ``arise'' and ``kill'' the moon, reversing the usual drama in
which the moon, or night, kills the sun, or day. There is also a second
opposition, in the last eight lines, between Juliet's sun and the stars,
heightened by an impossible hyperbole: Romeo imagines that Juliet's two
eyes might take the place of two stars and, outshining them, turn night
into day. By analogy the dierence in their order of brightness is likened
to the dierence between daylight and articial light. None of these con-
trasting descriptions is obviously metaphorical, although, in calling Juliet
`the sun', Romeo is using a metaphor (as well as in his use of `moon' for
Rosaline). The point of the multiple contrasts between the sun, moon,
lamp, daylight, and other elements is rather to set up an exemplication-
schema composed of the sun and these other objects. What `the sun'
expresses metaphorically depends on what the sun, relative to the schema
set up in the passage, exemplies.
Knowledge by Metaphorical Content 159

The context of the larger passage, and indeed of Shakespeare's whole

play, also circumscribes the range of presupposed features sorted by the
schema as a whole, although it does not enumerate or characterize the
eld explicitly. To begin with, our knowledge that the sun and moon-
exemplied features are to be ascribed to human agents (in the latter case
signaled by the relative wh-pronoun `who is already sick . . .') excludes a
range of inappropriate features, preselecting one potential space of fea-
tures. These f-presuppositions function in a way similar to Tversky's
process that selects and extracts the relevant features to be put to the
matching test for a similarity judgment. However, at a second stage, we
again look at the literary context of (1) to identify, from among properties
appropriate to humans, those features exemplied by the sun in the
schema: We look at the presuppositions of the play about the personality
of Juliet, at what makes her Romeo's love and ``lady,'' at the presupposed
features that make her ``fairer'' than all her rivals such as her ``vestal''
purity and chastity, and, most important perhaps, at all the features that
seem to draw the envy of the moon, Rosaline, and of Juliet's other peers.
So, to interpret the metaphor `is the sun' in (1), its metaphorical char-
acter (or the character of the underlying metaphorical expression) directs
us to the actual p- and f-presuppositions that constitute its context of in-
terpretation. These are p-presuppositions about the other elements in the
schema relative to which the referent of its literal vehicle exemplies
what it does and about the range of features sorted by the schema, and
f-presuppositions about Juliet and possibly other elements in the schema
that rule out inappropriate features, both for generic and specic reasons.
Notice that these presuppositions need not (although they might) include
any that specically, explicitly state the feature presupposed to be exem-
plied by the sun in the context. Rather the only way to identify the
complex of properties exemplied by the sun in the context may be by
way of the description `the complex of properties exemplied by the sun
relative to the schema hx; y; z; . . .i'. And indeed the complex of properties
may be one that has no simple adequate linguistic expression. As I'll
argue in section VI, the property might be expressible by us only by rela-
tion to the sample, relative to the specic schema that obtains in the con-
text; we describe the schema which, in turn, shows us what it samples.
Our grasp of the property itself may lack the abstractness that is a pre-
condition for a fully conceptualized linguistic expression to express it,
an expression that would mark the fact that we can determine whether
the property is possessed across contexts by arbitrary objects. But these
strictures on our verbal, linguistic, or conceptual ability to express the
160 Chapter 5

property need not imply that the property itself, or the content of the
metaphor in that context, is necessarily indeterminate or vague.
As I noted in our discussion of sampling, we cannot generalize from its
necessary and/or sucient conditions to exemplication in general. If that
is true even of relatively uncomplicated exemplication systems, how
more so of the subtle, complex exemplication involved in literary texts of
the kind we have been examining. There are no general instructions
how to determine what is exemplied, and discovery of what something
exemplies may take ``time, training, and even talent,''21 but especially
imagination. As our discussion in the previous paragraphs hopefully
illustrates, discovering what is exemplied is often the result of balancing
out our various p- and f-presuppositions, adding, subtracting, and can-
celing presuppositions, often without arriving at one round gure as the
result. Yet, the lack of rules or instructions and the trouble and eort
necessary to state denitively what is exemplied on an occasion should
not lead us to conclude that nothing is exemplied (and hence that noth-
ing is metaphorically expressed). As all of us know from our experiences
in interpreting original, challenging metaphors, especially in literary texts,
these sorts of diculties and complications are endemic to the enterprise.
The point of a theory of metaphor (or at least ours) is not to furnish rules
that eliminate this hard work, or even make it easier, or that make it
possible to grind out interpretations; rather the theory aims to specify the
parameters and factors knowledge of which will enable us to understand
why the practice and process of interpretation is (and ought to be) di-
cult. Two of these parameters are the schema and eld of features with
which exemplication varies.
To illustrate the context-specicity of these parameters, contrast the
above passage from Romeo and Juliet with another set of `sun' metaphors
based on a dierent exemplication-schema sorting a dierent eld of
features. Shakespeare frequently uses `sun' metaphors in his historical
plays; in the following passage, the Welsh Captain and Salisbury describe
the end of Richard II's reign:
Cap. 'Tis thought the king is dead; we will not stay.
The bay-trees in our country are all wither'd,
And meteors Mthat[`fright'] the xed stars of heaven;
The pale-faced moon Mthat[`looks'] bloody on the earth,
And lean-look'd prophets whisper fearful change;
Rich men look sad and ruans dance and leap,
The one in fear to lose what they enjoy,
The other to enjoy by rage and war:
These signs forerun the death or fall of kings.
Knowledge by Metaphorical Content 161

Farewell: our country men are gone and ed,

As well assured Richard their king is dead.
Sal. Ah, Richard, with the eyes of heavy mind
I see thy glory Mthat[`like a shooting star']
Mthat[`Fall to the base earth from the rmament'].
Mthat[`Thy sun'] Mthat[`sets'] weeping Mthat[`in the lowly west'],
Witness Mthat[`storms'] to come, woe and unrest;
Thy friends are ed to wait upon thy foes,
And crossly to thy good all fortune goes.
(Richard II, II, iv; my italics)
The exemplication-schema to which the sun belongs in this passage is
markedly dierent from that of Romeo and Juliet. Here the sun is not one
among a set of sources of light``fairness''but one among a class of
celestial ``signs'' that pregure crisis and tragedy, the ``death or fall of
kings'': ``meteors,'' ``wither'd bay-trees,'' the ``xed stars of heaven,''
``moon,'' ``earth,'' ``storms to come,'' and ``a shooting star.'' Whereas
Romeo's sun is rising in the East, the sun of this passage is setting in the
West. The range of features sorted by the schema are predominantly
negative: fear, sadness, the anxiety and insecurity of lawlessness; states of
ephemeral greatness, aging glory, slow, inevitable decline, imminent di-
saster and crisis. Relative to this schema, the (setting) sun exemplies (and
thus the metaphor ``thy sun'' expresses) (declining) glory, (lost) authority,
and insecurity. Again, the passage does not explicitly enumerate these
presupposed features. Instead it depicts a scene, or scenario, in which the
exemplifying elements in the schema display for us the properties they
exemplify.22
The properties (presupposed to be) exemplied by the sun (and then
metaphorically expressed by `the sun') in these two passages are highly
specic to their respective schemata. The author deliberately designs and
sets up the schemata, which, for all we know, hold only within these pas-
sages. In contrast, normal and stereotypical features, as we said in chapter
4, are associated with expressions (hence with literal vehicles of meta-
phors) invariantly across contexts, hence regardless of schemata to which
their referents belong. Although there are no strict criteria to distinguish
exemplication-metaphors from, say, metaphors based on normal notions
or stereotypes, where the schema is more or less explicit as in these ex-
amples, we might also use its degree of schema-relativity as one way of
diagnosing its metaphorical kind. Where the schema is not explicit, there
may also be default features that a given object exemplies. Out of con-
text, `the roses of her cheeks' may sample fragrance, pinkness, and soft-
ness, not thorniness or deep red.23 Similarly, animal metaphors for
162 Chapter 5

character-traits (e.g., `fox' for sly, `gorilla' for violent and dangerous, etc.)
or occupational or craft metaphors for manners (`butcher' for messy and
destructive, `surgeon' for careful and precise) exemplify relative to more
or less contextually xed schemata that sort contextually xed ranges of
features. But the less context-specic and more widely shared the schema
of exemplication, or where the metaphor has default interpretations, the
more dicult it is to distinguish a metaphor of exemplication from one
that draws on presuppositions concerning normal or stereotypical
notions. Who is to say, when Wittgenstein remarks
(2) The works of the great masters are suns which rise and set around
us. The time will come for every great work that is now in the
descendent to rise again24
whether `suns' is a metaphor that exemplies cyclical rise and decline, or
whether it simply builds on presuppositions that belong to our normal
notion of the sun?25
The larger issue raised by this question concerns the degree to which the
metaphorical interpretation of any single expression is necessarily relative
to a schema, network, system, or family of expressions to which it be-
longs. In recent years this issuethe systematicity of metaphorhas
become a leitmotif among philosophers, linguists, and psychologists.
Nelson Goodman, perhaps the rst to press the point, argues that in
metaphorical transfer, ``a label functions not in isolation but as belonging
to a family''; that it is always ``a label along with others constituting a
schema [that] is in eect detached from the home realm of that schema
and applied for the sorting and organizing of an alien realm.''26 Good-
man's own argument for the family-relativity of metaphor is rooted in his
general theory of predication, which he treats as a form of categorization
or classication, the basic cognitive activity in his lights. Since all classi-
cation is relative to a set of either implicit or explicit alternatives, the
predication `a is an F' is also a matter of classifying a as an F rather than
as a G, H, etc. The metaphorical application of a predicate is, in turn, no
dierent from other types of predication; hence it too is relative to an
assumed schema of alternatives, hence relative to a family. We need not
saddle ourselves with Goodman's theory of predication and, to be sure,
most contexts are not as explicit as our Shakespearean speeches in enu-
merating the members of the schema and the eld of features relative to
which a given metaphor should be interpreted. However, if exemplica-
tion is an underlying mechanism by which a signicant class of metaphors
Knowledge by Metaphorical Content 163

is interpreted, the fact that all exemplication is relative to a schema

would oer independent support for Goodman's insight.
There is also indirect evidence for Goodman's insight from the prac-
tices we employ in interpreting metaphors whose exemplication sche-
mata are not explicit. The rst step we typically take is to recover such a
schema, guided by the principle that what is metaphorically expressed is a
feature that some referent in the context, relative to a schema, exemplies.
To interpret the metaphor, we therefore enrich the context by adding such
a referent with its schema to our presuppositions. Consider, for example,
Hamlet's statement, ``What should such fellows as I do crawling between
heaven and earth?'' (Hamlet II, 2, 129). The statement could, to begin
with, be interpreted literally in its context. Yet, as I. A. Richards remarks,
it also has a denite metaphorical interpretation: ``[T]here is an unmis-
takable reference to other things that crawl, to the motions of foul insects,
to vermin, and this reference is the vehicle as Hamlet, or man and his
ways, are the tenor'' (Richards 1936, 119120). The force of the meta-
phor `crawling', Richards continues, ``comes not only from whatever
resemblances to vermin it brings in but at least equally from the dier-
ences that resist and control the inuence of their resemblances. The im-
plication there is that man should not so crawl'' (ibid., 127). That is, other
metaphors personify their inanimate or non-human agents; this metaphor
dehumanizes or depersonies its agent by attributing features of insects or
vermin and, with them, a negative evaluation.
Suppose Richards is right: Suppose the kind of agent whose crawling
Hamlet attributes to himself here is that of an insect or vermin. The
question we still need to answer is: why that subject rather than something
else that crawlsbabies, soldiers under siege, snakes, other animals? How
do we, as interpreters, know that this is the relevant agent of crawling? In
our choice, we can see the inuence of exemplication as a mechanism of
generating interpretations. In interpreting the metaphor, we try to identify
a referent, or extension, for the expression (e.g., `crawl') that exemplies a
property or kind of property we think appropriate for the subject of the
metaphorical statement. Our schema of alternatives might consist, then,
of the various things that crawl and a correlated set of features those dif-
ferent kinds of crawling things sort. Thus the crawling of snakes exem-
plies deceit; the crawling of babiesimmaturity, innocence, clumsiness,
perhaps lack of control; of soldiersstealth and surprise in attack. None
of these features would be appropriate to express of Hamlet (the referent
of `I') in this context; for it is f-presupposed in the context that Hamlet
164 Chapter 5

loathes himself and mankind in a t of misanthropy. So, what is (appro-

priately) expressed is a kind of repulsive degradation that only the crawl-
ing of insects exemplies. Hence we ll in the context to make the
crawling of the agent insect- or vermin-like, and the ``insectication'' of
the subject in turn brings in its train the other associations Richards notes.
Here exemplication guides the way we ll in, or transform, the context in
order to yield a metaphorical interpretation.27
If we think of Goodman's thesis of the family-relativity of metaphorical
interpretation in terms of exemplication-schemata, we can also clarify an
obscure point in his conception of metaphor as transfer. According to
Goodman, when we use, or apply, any one expression metaphorically,
what is transferred is ``a whole set of alternative labels, a whole apparatus
of organization [that] takes over new territory'' (Goodman 1976, 73).
That is, the whole schema in which the given expression functions is
metaphorically ``transferred'' to the alien realm. But does that mean that
each member of the schema is ``transferred'' in exactly the same sense as
the individual metaphor used on the occasionthat each element in the
schema is itself interpreted metaphorically on the occasion? In our ex-
ample from Romeo and Juliet, `moon' as well as `sun' is interpreted
metaphorically. But do all the terms for the other elements in the
exemplication schema, for example, `the brightness of a cheek', `lamp',
`daylight', and `night' have metaphorical interpretations? (What would
they be?) In my own quotation of Shakespeare's passage, I identied only
`sun' and `moon' (and their governing verbs `arise' and `kill') as meta-
phorical expressions, because only those expressions obviously undergo a
change of content, a change in their respective contributions to the truth-
conditions of their utterance. On the other hand, if not all elements in the
schema are in fact metaphorically interpreted, how should we understand
Goodman's idea that every time we use a single metaphor, its whole
family or schema is transferred?
I want to suggest that, in putting forth his thesis, Goodman's real con-
cern is with the individuation of a metaphor. What counts as one meta-
phor and what counts as two, and by what criteria? When do we have
the same and when dierent metaphors? I have touched on this question
earlier.28 We obviously cannot individuate metaphors by the number of
words used (one word, one metaphor); in this sentence from a Time
Magazine lm review, all the words in the italicized phrase (which is not
an idiom) constitute one metaphor:
(3) [The movie That's Dancing!] is lmed dance with one leg tied behind
its back. (Time, Jan. 28, 1985)
Knowledge by Metaphorical Content 165

Likewise, it is arguable that in Conrad's statement

(4) Into that narrowed circle furious seas leaped in, struck, and leaped
out
it is all three italicized verbs that together constitute one metaphor. (In
(4), `furious' is, of course, also metaphorical, and metaphorically related
to the italicized verbs, but it is a distinct metaphor.) We also cannot
individuate a metaphor by surface phrasal constituency; on one reading of
Davidson's example (discussed earlier in ch. 4, fn. 35):
(5) Tolstoy was a great moralizing infant,
the metaphor is discontinuousonly `a great infant' is metaphorical
violating the phrasal structure.29 Similarly, it is arguable that Luke's
proverb
(6) Physician, heal thyself (Luke, iv, 23)
is one metaphor, despite the fact that `physician' and `heal' belong to
syntactically distinct surface phrases.30 In short, we cannot individuate
metaphors either simply by word units or by surface phrasal units.
In terms of my theory, we can distinguish at least three dierent criteria
for the individuation of a metaphor; Goodman's intended criterion is the
third of these. I would also argue that, depending on our other interests
and purposes, we employ each of these criteria at one or another time.
The rst criterion is sameness of character; thus the three `sun' metaphors
in Romeo and Juliet, Richard II, and the Wittgenstein passage, or any of
the other pairs of metaphors cited in chapter 1, examples (2)(10), such as
(repeated here)
(7) Life is a bubble
(8) The earth is a bubble
count as one metaphor since each expression interpreted metaphorically is
a token of one metaphorical expression type, for example, `Mthat[`is a
bubble']'. It is in this sense that we say that a (i.e., one and the same)
metaphor can have dierent interpretations (contents) in dierent con-
texts. Formally:
(i) Two occurrences of F and c are the same metaphor i
{Mthat[F]} {Mthat[c]}.
The second criterion requires sameness both of character and of content
in their respective contexts, that is, relative to their respective context sets
of presuppositions. By this standard, (7) and (8), as well as each of our
166 Chapter 5

`sun' metaphors, each count as a dierent metaphor, since their respective

contents in their respective contexts are dierent. Here we individuate the
metaphor by what is metaphorically expressed (in the context) as well as
its metaphorical form of expression. Formally:
(ii) Two occurrences of F and c are the same metaphor i
(a) {Mthat[F]} {Mthat[c]} and
(b) in their respective contexts c and c 0 , {Mthat[F]}(c) {Mthat[c]}(c 0 ).
The third criterion for individuation of a metaphor takes into account
not only the character and content of the individual expression interpreted
metaphorically (or its corresponding metaphorical expression) but also
(where there is one) the family (say, exemplication-schema) of the literal
vehicle of the metaphor. This criterion is also, I propose, what Goodman
is trying to capture with his thesis about family transfer. To appreciate the
force of the intuition that drives it, let's look again at the quotation from
Romeo and Juliet. I have identied only a few words (`sun', `moon',
`arise', `kill') as metaphors, as I said, based on their content, that is, truth-
conditions, in the context. But is it really clear that the metaphors begin
and end with these individual expressions? Suppose we x the content of
the metaphor `is the sun' in its context. `Juliet is the sun', interpreted
metaphorically, says that she is far `fairer' than any rival, so far fairer that
her competitors are virtually eliminated from the competition. She `kills'
and `shames' them in competition. (Again, this is not meant to be any-
thing like a denitive interpretation, merely exemplary of what a xed
content could be.) Yet, however we x it, this content of the metaphor
does not exhaustfar from itall the information or cognitive signi-
cance it conveys. In chapter 7, I'll return to the theoretical nature of this
additional information or cognitive value of a metaphor and its dieren-
tiation from its content in a context. For now it is sucient to observe
that there is clearly much more metaphorically conveyed by the various
ways in which the sun is said to be related to the other elements in its ex-
emplication schema, for example, the way in which the sun's bright-
nessviolently?`shames' the stars and `kills' her rivals, especially the
envious moon. Or the contrast between her natural `daylight' and the
articial, indirect light of a `lamp'. Or between her sustained brightness
and the twinkling of the stars. This third criterion tries to incorporate
all these added dimensions of signicance, shown or displayed by the
exemplication-schema, into its conditions of individuation.
The other terms in the exemplication-schema may be used (only) lit-
erally, but passages like this challenge a neat line of separation between
Knowledge by Metaphorical Content 167

the metaphorical and the literal (which is not to deny the distinction), that
is, a simple division between expressions used on an occasion that are to
be identied as metaphors and those to be typed as literal. Furthermore,
although `is the sun' is interpreted metaphorically, it must also have, in
some sense, its literal meaning. For the literal vehicle of the metaphor,
together with its schema, is what generates the m-associated presup-
positions (about what is exemplied by the sun) out of which we produce
the metaphorical interpretation. (Recall our discussion in ch. 2 of the
diculties in demarcating this sense of ``have''; I shall return to the
appropriate sense in ch. 6.) We could then include all other members of
the schema (under their literal characters) in our criteria of individuation
for the metaphor. For the character, and resulting interpretation, of the
metaphorical expression depends on its literal vehicle, and its literal vehi-
cle contributes to the metaphorical interpretation by way of its contrast-
ing relations to the other members of the exemplication-schema.
Another way to focus on this third criterion would be to contrast
the `sun'-metaphor (in this extended sense) determined by its whole
exemplication-schema with a second metaphor (in the same extended
sense) in the same passage. In lines 1214 Romeo appeals to a second
metaphor based on the widely used Elizabethan metaphor that people (or
human faces) are books (or, in this case, oral speech):
She ?Mthat[`speaks'], yet she ?Mthat[`says'] nothing; what of that?
Her eye Mthat[`discourses'], I will not Mthat[`answer'] it.
I am too bold, 'tis not to me she Mthat [`speaks'].31

I count these, the extended `sun'-metaphor and the `face'/`speech' meta-

phor, as two dierent metaphors: Each set of expressions constitutes one,
internally coherent schema of metaphorical interpretation, and they are
dierent metaphors because, as schemata, they are independent of each
other.32 Again, it is not entirely clear where to draw the line between the
metaphorical and the literal. We can x the content of the individual
metaphor, but when we try to spell out its undeniable additional cognitive
signicance, it is impossible not to appeal to what is conveyed by its
whole schema of interpretation, including its literal elements.
Yet another aspect of the schematic nature of a metaphor suggests that
it be individuated by a larger, more systemic unit. Consider the following
passage from Time Magazine (cited earlier in ch. 1):
Propelled by the engine of the postwar Wirtschaftswunder, the capitalist Federal
Republic of Germany is a sporty blond racing along the autobahns in a glittering
Mercedes-Benz. The Communist German Democratic Republic, bumping down
168 Chapter 5

potholed roads in proletarian Wartburgs and Russian-built Ladas, is her homely

sister, a war bride locked in a loveless marriage with a former neighbor. (Time,
March 25, 1985 )
This passage contains at least two exemplication schemata (each with
two explicit elements): `a sporty blond' vs. `her homely sister, a war bride
locked in a loveless marriage with a former neighbor' and a sequence of
carsMercedes-Benz, Wartburgs, and Ladas. Numerous expressions are
individually interpreted metaphorically. Yet if one (somehow) failed to
see that this whole passage, with all its rich parts and complex inter-
relations, constitutes one metaphor, some kind of unity as a metaphor, she
would miss a crucial part of its interpretation and power.
On my theory, the schema is not itself strictly speaking part of the
metaphor; it is part of the context necessary to generate the m-associated
presuppositions about what the referent of the literal vehicle for the met-
aphor exemplies. So, according to this third criterion, we might say that
we individuate the metaphor by its character and content-in-its-context
which collapses onto the second criterion. Formally, however, we might
distinguish this third criterion from the second by building its mode of
generation (including its contextual schema of exemplication) into our
representation of the content. Recall that back in chapter 3, section III, I
suggested a way of representing logical but nonpropositional structure in
Russellian (or referential) propositions. Although the propositional con-
stituent for a metaphorical expression Mthat[F] is, say, a property P, we
can display the character-istic information that yields P as its value in its
context c (or, more specically, context set) by representing P as the value
of the character of the metaphorical expression in its context: that is, as
{Mthat[F]}(c). By the same token, we can make explicit in our represen-
tation all the contextual parameters that yield P, given the character of
the metaphorical expression. Let the context set csi for the metaphorical
expression Mthat[F] contain, among other presuppositions, the presup-
position that F (or Fs) belong to an exemplication-schema E. We
can represent that presupposition as hF; Ei. So csi h. . . ; hF; Ei; . . .i.
Thus
(iii) Two occurrences of F and c are the same metaphor i
(a) {Mthat[F]} {Mthat[c]} and,
(b) in their respective contexts c and c 0 ,
where c h. . . ; hF; Ei; . . .i and
c 0 h. . . ; hc; E  i; . . .i, {Mthat[F]}(c) {Mthat[c]} (c 0 ), &
(c) c c 0 , i.e., h. . . ; hF; Ei; . . .i h. . . ; hc; E  i; . . .i.
Knowledge by Metaphorical Content 169

In other words, not only must the same metaphors have the same char-
acters and contents; their contexts, in particular where they involve ver-
bally explicit schemata of interpretation, must also be the same.
Goodman's family or network that is transferred in metaphor is spe-
cically the set of alternative predicates by which a realm of objects is
sorted into categories. Structural linguists sometimes call such networks
contrast sets.33 Another example of a contrast set is my notion of an ex-
emplication schema. But this is not the only kind of network or set of
systematically related expressions that enters into metaphorical interpre-
tation. In the next sections, I shall turn to two others. These networks
supplement rather than compete with the exemplication schema for
metaphors of that kind.

IV Thematic and Inductive Networks

To borrow a metaphor from Quine, metaphors do not face the tribunal of

interpretation one by one but as a corporate body. Indeed there are dif-
ferent kinds of corporate bodies into which metaphors organize them-
selves. Some of these other systems or networks have been discussed by
recent authorsincluding George Lako, Mark Johnson, Eve Sweetser,
Mark Turner, Eva Kittay, Adrienne Lehrer, Lynne Tirrell, and Roger
White.34 The details of their respective accounts vary widely, and they
draw on dierent kinds of metaphors to support their respective theories.
I shall not review them in detail, nor shall I argue for one to the exclusion
of others. As I have said, there does not seem to be a single network or
one set of systematic relations that explains, once and for all, how meta-
phors work. Under the rubric of a single semantics, there can be dierent
kinds of networks subsumed as dierent aspects of context. Semantic
orthodoxy is compatible with broad, variegated pragmatic heterodoxy.
However, the more dierent kinds of networks we can identify, the better
feel we shall have for the multiple forces acting from dierent directions
that shape metaphorical interpretations.
Two additional networks or metaphor-systems, which I shall now dis-
cuss, interact in various ways, but they dier in one important respect.
The rst network, the thematic scheme of a metaphor, supports pre-
dictions as do the semantic constraints of character; our knowledge of this
network is also linguistic: semantic or lexical. The second kind of net-
work is built on associations of dierent degrees of inductive strength.
This network of inductive associations (for short, inductive network)
does not furnish us with the means to lay down law-like claims about
170 Chapter 5

metaphorical interpretations, but the fact that it does not support pre-
dictions makes it no less important. If we fail to see that a number of
metaphors belong to one inductive network, we will still have missed a
signicant generalization.
The idea of a thematic metaphor scheme, or a system of thematically
related metaphors, is not new. In Eva Kittay's (1987) version of semantic
eld theory, the notion corresponds roughly to the set of syntagmatic
relations of a word, its relations to the words that constitute its linguistic
context, with which it either obligatorily or optionally collocates.35 For
example, the syntagmatic eld related to the verb `to sh' (in Kittay's
terminology: the conceptual domain of shing) contains phrases for an
agent (`sherman'), patient (`sh'), instrument (`with rod/line/bait', `line
and hook', `net', `angling reed'), goal (`catching a sh'), and location (`in
the water', `in the crystal brooke', `river'). In metaphor, this eld of words
structured according to these grammatical relations is transferred as a
whole to another conceptual domain, for example, courtship, as Kittay
([1987], 263275) illustrates with Donne's ``The Bait.''
In current theoretical linguistics, the notion of a thematic relation or
role corresponds roughly to the role identied by a syntagmatic rela-
tion.36 Thematic relations concern the number and grammatical type of
(phrasal) arguments required by the verb of a sentence (where the verb
refers to an action, process, or state) and the semantic and syntactic rela-
tions the arguments bear to the verb. Thematic roles typically include
Agent (the initiator or doer of an action), Goal (that toward which some
concrete or abstract event moves or changes), Source (that from which
some concrete or abstract event moves or changes), Location (the place
where something is), Experiencer (the individual who feels or experiences
the event), Recipient (that which receives, where there is change of pos-
session), Instrument (the means with which an action is performed),
Benefactive (the one for whose benet an event or action takes place),
Percept (what is experienced or perceived), Patient (what undergoes an
action), and Theme (what moves or changes in any abstract or concrete
motion or change). On at least some theories a single phrase can also bear
more than one thematic role. These relations and roles are not determined
by the structural position of a phrase in a string; rather they are deter-
mined by the meaning of the verb. That is, we know the thematic roles to
be assigned in a given sentence and the role to be borne by each phrase
(only) by knowing the meaning of the dominating verb. However, this
knowledge is not necessarily of each verb one by one. Verbs also organize
into classes, each of which has its respective lexical and syntactic proper-
Knowledge by Metaphorical Content 171

ties according to which its members interact with the syntax in predictable
ways. So, in addition to knowledge of the idiosyncratic meanings of indi-
vidual verbs, a speaker also knows the properties of these verb classes and
the membership of each class. All this knowledge of verb-assigned the-
matic roles is linguistic, hence in contrast to our extralinguistic knowledge
of exemplication-schemata. In other words, the signicant point for us is
not (as it is for Kittay) that the thematically or syntagmatically related
noun phrases, given their standardized names (e.g., `sh', `sherman'),
belong to one semantic eld corresponding to a unied conceptual do-
main, but rather that the noun phrases are linguistically determined argu-
ments for verbs with certain general properties of meaning.
To see how the thematic network of a metaphor aects its interpreta-
tion, let's focus on verbs.37 When one uses a verb metaphorically, as
in (9):
(9) Sleep . . . knits up the raveled sleeve of care (Macbeth II, ii)
the idea is that not only the verb (`knits up') is ``involved'' in the meta-
phorical interpretation, but also the noun phrases in the sentence that
play the thematic roles it assigns. First, `knits up', whether interpreted
metaphorically or literally, requires an Agent (in the subject position) and
a Patient or Theme in object position. Hence the unacceptability of, for
example,
*(9a) Sleep knits up;
*(9b) Knits up the raveled sleeve of care.
Thus the metaphorical interpretation of the verb preserves its thematic
structure. Second, while `sleep' still (literally) refers to sleep, its Agent
thematic role personies or `agent-izes' it; by virtue of its thematic role of
Agent, `sleep' acquires whatever features are necessary to be an agent.
Likewise, the patient or theme of the verb `knits up' (metaphorically as
much as literally) must be an appropriate kind of object. `Knits up' (lit-
erally) means `to compose or repair by knitting, that is, by tying by fas-
tening with a knot (as in weaving)'. Therefore, its patient must be the sort
of thing that has disconnected or thread-like parts that are composed or
repaired by knotting, interlacing, or intertwining them. This, in turn,
excludes as patient or theme any noun phrase referring to partless kinds of
things, such as `care'; hence the unacceptability of
*(9c) Sleep knits up care.
In this respect, contrast (9) with (10):
172 Chapter 5

(10) He bathes in gladness.

The verb `bathe' requires an agent (if `bathes' is an action) or experiencer
(if `bathes' is a state) and (optionally) a location (here the prepositional
phrase `in X').38 However, the location (or instrument) argument for
`bathe' can either have as its ``head'' a mass noun (`water', `oil', `mud') as
in (10) or a count noun ` (`a bathtub', `a pool of water', `a puddle') as
in (10a):
(10a) He bathes in a tub/pool/sea of gladness.
Either (10) or (10a) is acceptable. In contrast, (9) does not admit these
options. `Care' must be incorporated into a complex noun phrase whose
referent is of a kind that has parts that can be tied or fastened together,
for example, `the raveled sleeve of care'.
To be sure, the requirements of the thematic roles necessary for a verb,
together with these additional semantic features, do not uniquely deter-
mine a single noun phrase. Any candidate that satises the relevant con-
ditions would be equally acceptable. Shakespeare might have written `the
raveled sweater of care' or `the tangled loose threads of care', for exam-
ple. It is in choosing or discriminating among these alternative possibil-
ities that the true master of metaphor demonstrates his skill. However,
what is not up to the metaphor-maker is that there must be some noun
phrase with these semantic properties to ll that thematic role.39 The
thematic network of a (verb) metaphor thereby both implicates other
expressions in its interpretation and constrains their involvement.
It should be emphasized that these network relations can be captured
only in terms of the character of the metaphor, not in terms of its content.
Thus the content of (9) is roughly:
(9d) Sleep restores (repairs, composes) the calm and composure of
confused, exhausted, weary people.
Since `restores', `composes', and `repairs' do not require a theme or
patient with parts,
(9e) Sleep restores (repairs, composes) calm (to the weary and confused)
is also acceptable. It is only at the level of the character of (9), that is, the
character of
(9f ) Sleep Mthat[`knits up'] Mthat[`the raveled sleeve of '] care,
that we can represent the kind of semantic structure relevant to these
thematic roles that explains why the metaphor takes the form it has.
Knowledge by Metaphorical Content 173

Attending to the thematic argument structure of a metaphor can also

inuence the interpretation we give it. Consider Aristotle's chestnut:
(11) The ship ploughs the waves.
On the face of it, the subject (`the ship') is agent and the direct object (`the
waves') is patient, analogous to the literal
(12) The oxen plough the eld.
This is also how the literary scholar Hugh Kenner reads the metaphor
whose content, he says, has to do with the ``similarity in two dissimilar
actions: the ship does to the waves what a plough does to the ground.''40
But it is not at all clear what it is that a ship does to waves that resembles
what oxen do to a eld when they plough. Instead I would suggest a dif-
ferent metaphorical interpretation of (11) based on a dierent thematic
structure: The motion of the ship through the waves is in the manner of a
plow, rising and descending with the waves as a plow rises and descends
with the furrows in the ground. That is, the metaphor expresses the eect
of the waves on the motion of the ship, rather than, like (12), the eect of
the ship on the waves. On this reading, the subject is not the Agent but the
Theme, while the object noun phrase is the Source and/or Location of the
action. `Ploughs', in turn, is a verb of a manner of motion.
The moral to be drawn from this example is, again, that it is not always
enough to interpret a single expression metaphorically; we must also take
into account the other expressions that discharge its thematic roles, the
whole system of relations between the words used in the context. Only
when we interpret, in this example, the main verb as part of its whole
network of thematically related words can we arrive at its correct meta-
phorical interpretation.
Thus far we have discussed only verbs in connection with thematic
networks because only verbs directly assign arguments with thematic
roles. One might try to extend this story to nominals derived from verbs,
for example, to `sherman' which is derived from (verb) `sh', such that
the thematic roles assigned by the verb are inherited by the noun. Where,
for example, there is an agent signied by `sherman', it would be implied
(even if it is not verbalized) that there is a patient (signied by `sh' or
some subordinate class, e.g., `trout' or `salmon'), location (`water', `lake',
etc.), and instrument (`hook', `bait', `line', `net', etc.). And to extend the
idea even to underived nominal phrases, one might argue that many
common nouns or noun phrases have characteristic or ``normal'' actions
associated with them; for example, `the sun shines/warms/blazes'. Given
174 Chapter 5

these associated verbs, the thematic roles they assign might, in turn, be
inherited by the respective noun, yielding a network of associated roles
that can be assigned to still more nouns or noun phrases. To spell this
out, I now turn to the second kind of network based on inductive
associations.
As several writers on metaphor have noted, all the metaphors in the
following lines in T. S. Eliot's ``The Love Song of J. Alfred Prufrock''
presuppose an enthymematic metaphor: `fog is a cat'.
The yellow fog that rubs its back upon the window panes,
The yellow smoke that rubs its muzzle on the window panes,
Licked its tongue into the corners of the evening,
Lingered upon the pools that stand in drains,
Let fall upon its back the soot that falls from chimneys,
Slipped by the terrace, made a sudden leap,
And seeing that it was a soft October night,
Curled once about the house and fell asleep.
(my italics)
Tanya Reinhart (1970, 391) makes this point, using I. A. Richard's ter-
minology, by calling a cat (or another animal with similar behavior) the
metaphorical vehicle. Lynne Tirrell (1989, 2628) proposes that the
implied `fog is a cat' metaphor functions as an ``antecedent,'' or govern-
ing, metaphor (the ``metaphor-proper,'' in her words) for the extended set
of (italicized) metaphorical expressions, roughly in the way that a noun
phrase serves as antecedent for co-referential pronouns in anaphoric
chains. What both authors have in common is the idea that the con-
stituent metaphors in the passage belong to one interrelated complex, one
system or network. Someone who interprets each expression metaphori-
cally but fails to recognize that they are part of one network misses an
important fact about them as metaphorseven if he assigns the same
interpretation to each individual metaphorical expression as the one who
recognizes their network-structure. In chapter 7 I'll return to the cognitive
signicance of the network-structure carried by the characters of its
members. Here I want to look at the status of the connections between the
predicates in these networks or, more precisely, the relations between the
properties they express.
One way to explain the connections between the italicized phrases in the
above ``Prufrock'' passage would be to take them to be deductively re-
lated. Each of the phrases belongs to one `fog is a cat' metaphor because
our concept of a cat can be explicated or analyzed by or decomposed
(exhaustively or not) into the following set of meaning postulates:
Knowledge by Metaphorical Content 175

(i) (x) (x is a cat ! x is yellow)

(ii) (x) (x is a cat ! x rubs its back)
(iii) (x) (x is a cat ! x makes sudden leaps)
(iv) (x) (x is a cat ! x licks its tongue)
(v) (x) (x is a cat ! x curls up to sleep), etc.
But the obvious problem with this approach is that it is not at all clear
which properties to include in our concepts. To make the approach
descriptively adequate to capture all the kinds of cat-scheme predicates
that can be used in one such extended metaphor, it would be necessary to
include every property that is empirically true of a cat in its concept (and
perhaps even more, such as false stereotypical properties). But once we
include all such properties, every change in our (empirical) knowledge
about cats would constitute a change in our concept of a cat. On the other
hand, if we restrict the properties we take to be constitutive of the con-
cept to some privileged subset, we need some criterion to distinguish
those constitutive properties from all others. For lack of a good criterion,
a sharp division runs the dangers of an unprincipled analytic-synthetic
distinction.41
A more plausible explanation of the fog-is-a-cat network is that it is
inductive in character. That is, all the properties are inductively related to
our experiences of cats, but to dierent degrees of conrmation or prob-
ability. So, `x is licking its tongue' and `x is curling up' can be more
strongly inductively inferred from `x is a cat' than `x is rubbing its back'
or `x is lingering', and all of these more than `x chases dogs'. Inversely,
from `x is yellow', `x licks its tongue', and `x is curled up to fall asleep' we
can infer, though again with dierent degrees of likelihood, `x is a cat'.
Here the nature of the inference from each of the italicized phrases, and
from all of them as a group, to `x is a cat'as the ``governing'' or ``con-
trolling'' metaphorwould not be deductive but that of an inference to
the best explanation.
All the cat-associations mentioned above are more or less community-
wide and socially shared, rather than personal and idiosyncratic. Indeed
they are not just relative to onethat is, ourcommunity but are more
or less constant across communities. And they are established rather than
novel associations. Other inductively based extensions might be estab-
lished only relative to a given community, and yet others may be novel
rather than established. Had Eliot metaphorically described a fog as a cat
becauselike a cat for a community in which it is a sacred, awesome
176 Chapter 5

animalit is o-limits, feared, and terror-inspiring, the metaphor would

be foreign to our community. Had he described it as a sometime black cat
because it is an omen of bad luckwhen it crosses or blocks our paths
it would be more of a novel than extended interpretation of the metaphor,
yet native to our community rather than foreign. In any case, unlike the
linguistic connections established by thematic relations, all of these links
are based on inductive, empirical grounds.
In sum, all three of the kinds of networks or schemes we have dis-
cussedexemplication, thematic, and inductivebelong to the context
that generates the presuppositions from which metaphorical inter-
pretations are drawn. This is not to suggest that these three should
exclusively and exhaustively be taken to be the sources of metaphorical
presuppositions. On the contrary, given the ways they dier among
themselves, there is good reason to expect that there are more sources, a
variety of dierent contextual factors that generate metaphorically rele-
vant presuppositions, factors that cannot be systematically reduced or
even unied into one overarching system. This state of pragmatic aairs,
as I said earlier, stands in contrast to the genuine semantic constraints
that belong to our semantic competence in metaphorical character.
Before leaving the pragmatics of metaphorical interpretation, it would
be helpful to contrast our networks with those in the best-known account
that analyzes metaphors in terms of large-scale, systematic networks:
the theory of ``conventional'' or ``conceptual'' metaphor developed by
George Lako and his co-authors Mark Johnson and Mark Turner.42 In
what follows I shall refer to the theory of this school simply as ``Lako
et al.'' or as ``Lako '' (with no slight intended to the others).

V Lako et al. on Metaphor

Lako et al. have collected to date the largest corpus of data in the liter-
ature on metaphorical networks with which they have constructed a val-
uable taxonomy of metaphor systems. Although the idea that metaphors
work in networks (or families) is not original with Lako et al., until their
research no one, I think, had a full appreciation of the ubiquity and
variety of these families. Lako et al. have also drawn some very strong
conclusions about the philosophical signicance of their descriptive ma-
terialabout the bankruptedness of ``objective truth,'' classical ``West-
ern'' philosophy of language, epistemology, and metaphysics, and much
contemporary theoretical linguistics, as well as a slew of other theoretical
claims. These polemically charged claims have not only failed to convince
Knowledge by Metaphorical Content 177

philosophers; they have also had the unfortunate eect of discrediting the
authors' empirical theory in the eyes of many beholders. For reasons of
space, I shall not discuss their philosophical claims.43 But it would be
helpful to examine Lako 's metaphor networks to see whether they con-
stitute a genuine alternative to those we have discussed.
According to Lako et al., there are three fundamental ways in which
their ``contemporary'' theory of metaphor diers from ``classical theories''
for whom ``metaphor was seen as a matter of language, not thought''
(Lako 1993, 202):
1. On Lako 's view, metaphors are not linguistic expressions (or inter-
pretations) but cross-domain mappings in the conceptual system. One
domain, the source, is used to ``conceptualize'' a second, the target, (i) by
individuating the entities of the latter in terms of its own (source) entities
(sometimes indeed ``making,'' or constituting, entities in the target do-
main), and (ii) by ``sanction[ing] the use of source domain language and
inference patterns for target domain concepts'' (ibid., 207). A ``meta-
phorical expression'' is simply ``a linguistic expression (word, phrase,
sentence) that is a surface realization of such a cross-domain mapping''
(ibid., 203). To designate the mappings, Lako et al. use capitalized slo-
gans, for example, ARGUMENT IS WAR or LOVE IS A JOURNEY
or TIME IS A MOVING THING. However, the verbalized slogans
should not be mistaken for the true metaphors, that is, the conceptual
mappings and restructurings of the target domains. I'll refer to the cap-
italized clauses as metaphorical slogans.
2. On Lako 's view, metaphor is not restricted to ``novel or poetic lin-
guistic expression.'' Instead ``everyday abstract concepts like time, states,
change, causation, and purpose also turn out to be metaphorical'' (ibid.).
Because these metaphors are ubiquitous, automatic, and often commu-
nally shared in ordinary language, Lako et al. call them conventional
metaphors. Poetic metaphors are typically ``based'' on the same mappings
as the conventional metaphors but they are not automatic; instead they
are often original or novel, requiring eort to be understood. These dif-
ferences should not, however, obscure the basic fact that both conven-
tional and poetic metaphors are realizations of the same mappings.
3. The primary evidence for Lako that metaphor is ``conceptual'' rather
than ``linguistic'' are the many complex and systematically organized
networks of metaphorical expressions with which we talk about domains
or topics. For example, aspects of love relationships are expressed using
metaphors from the domain of journeys:
178 Chapter 5

Our relationship has hit a dead-end street. Look how far we've come.
We're at a crossroads. We'll just have to go our separate ways. We can't
turn back now. I don't think this relationship is going anywhere. We've
gotten o the track. We are spinning our wheels.
These are not unrelated, independent individual metaphors. Rather ``we
have one metaphor, in which love is conceptualized as a journey. The
mapping tells us precisely how love is being conceptualized as a journey.
And this unied way of conceptualizing love metaphorically is realized in
many dierent linguistic expressions'' (ibid., 209). Thus lovers correspond
to travelers, their relationship to the vehicle, their shared goals to their
common destination, and diculties in the relationship to impediments in
their way. Given these correspondences, ``inference patterns used to rea-
son about travel are also used to reason about love relations'' (ibid.). And
similarly for other topics like argument, which is conceptualized as war,
or time, conceptualized as money, and so on.
I (and most philosophers, I think) would agree with most of these
claims. Metaphors are not restricted to poetry; they run through all uses
of language. Metaphor is also not linguistic as opposed to conceptual if
that means that language is merely decorative or that it does not express
cognitive content. Metaphorical interpretations can not only express cog-
nitive content, including novel properties, that would not be expressed
literally; they also bear cognitive signicancethat bears on action as
well as knowledgeby way of their (metaphorical) character, as I'll
argue in chapter 7. Finally, I share Lako et al.'s conviction in the exis-
tence and importance of large-scale systematic networks to which meta-
phors typically belong that bear on their interpretation. Much of the work
done by my thematic schemes and inductive networks is done by Lako 's
``experiential gestalts,'' constrained by his Invariance Principle that
ensures that structural (many of them thematic) relations among the
expressions taken from the source domain continue to hold in their
metaphorical application to the target domain. For example, in the LIFE
IS A JOURNEY mapping, the source domain (JOURNEY) includes an
argument place for paths, sources/places of origin, and goals/destinations,
none of which is clearly found in the target domain (LIFE). A primary
eect of the mapping is to ``create'' and structure argument-places for
those thematic roles in our (metaphorical) language for the target domain.
Despite its descriptive riches, Lako et al.'s explanation still has serious
gaps. They do not explain why a particular metaphorical expression
ought to be subsumed under one rather than another mapping or how the
Knowledge by Metaphorical Content 179

proper level of mapping is decided.44 Even in the LOVE IS A JOURNEY

example, it is not clear why it is specically love rather than any rela-
tionship or indeed any goal-directed activity (of which love might then
simply be an instance) that is conceptualized as a journey. They also fail
to explain why, given a conceptual metaphor, only some but not other
``parts'' of the metaphor are mapped or, when interpreted, appropriate.
Indeed, despite their sometimes inated rhetoric, in their more sober
moments, Lako et al. acknowledge that their schemes and their struc-
tural relations are not predictive and that they only provide constraints on
the class of possible metaphors (ibid., 215).
The main dierence between Lako et al.'s account and mine concerns
the relation between context and the metaphorical networks.45 One small
expression of this dierence is that even though the cross-domain map-
pings are the real metaphors for Lako et al., the realizations of those
conceptual metaphors in language are always in linguistic expressions,
that is, linguistic types. On my account, metaphors are never expression
types per se but interpretations (or uses) of expression tokens in contexts,
although those interpretations may depend on the networks, or mappings,
to which the token belongs. Consider Lako et al.'s examples of the non-
metaphorical: sentences (i.e., types) such as `the balloon went up' or `the
cat is on the mat'. Yet, it is obvious that even these ``literal'' sentences or
expressions can, in some context with a little imagination, be used meta-
phorically. T. S. Eliot might have continued his famous depiction of fog
as a cat with `the cat is on the mat' or we might summarize the impact of
a stock market run on a vulnerable economy with `the balloon went up'.
Contra Lako et al., there are no expressions (types) per se that are either
metaphorical or nonmetaphorical.46
Lako et al.'s conception of metaphor as any cross-(conceptual) do-
main mapping is also both too broad and too general. It is too broad
because there are too many dierent kinds of context-independent
mappings, not all of which plausibly underlie metaphor specically. And
it is too general because even multiple mappings are inadequate to
account for certain dierences of metaphorical interpretation for which
it is necessary to appeal to the role of context. Let me illustrate both
shortcomings.
Among Lako et al.'s conceptual metaphors, or cross-domain map-
pings, is a large number in which various domains (birth, states and
changes of state, death, life, form, progress, and time) are conceptualized
as, or structured, using (thematic) relations and concepts drawn from the
domain of space and motion in space. Lako et al.'s primary evidence for
180 Chapter 5

the existence of these metaphor systems, or cross-domain mappings, is the

variety of linguistic expressions in context-invariant conventional lan-
guage that realize them. Thus people come into the world when they are
born and leave it or depart when they die, we go in and out of states,
progress is moving forward, covering ground, and so on. These expres-
sions, they claim, are all metaphorical and they are metaphorical because
they realize an underlying cross-domain mapping. This, however, is not
the only possible explanation. Ray Jackendo (1983) and Jerey Gruber
(1965/1976) have, as an alternative, argued for what they call the The-
matic Relations Hypothesis: that there is one neutral, primitive structure
that is used both for the analysis of spatial location and motion and for a
large number of event and state semantic elds (including temporal, pos-
sessive, identicational, circumstantial, and existential elds). The only
dierences that exist among the semantic elds that possess this one
structure consist in (i) the things assigned to the roles of theme and loca-
tion and (ii) the semantic values of the referential expressions. All other
predicates in the dierent elds can be analyzed or decomposed (though
not always exhaustively, allowing for idiosyncratic dierences) into the
same spatial primitives for motion (e.g., GO, REMAIN/STAY, FROM,
TO, TOWARD) and a small number of general primitives (e.g., CAUSE,
BE, LET). Their evidence for this hypothesis is lexical, as well as the
many syntactic regularities that recur across these elds.
In large measure, Jackendo-Gruber and Lako et al. are in agree-
ment. Spatial notions and their thematic relations, both of which consti-
tute systematically organized linguistic or conceptual networks, structure
our thought and talk in many other domains. Their disagreement is
whether the evidence warrants calling all this talk and thought metaphor-
ical. For Lako et al., the large number of expressions that linguistically
realize the cross-domain mapping is primary evidence for the ubiquity of
metaphor. Jackendo denies this for four reasons. I am sympathetic to his
conclusion, but his arguments do not, I think, make the case. The rst two
reasons can be dealt with quickly. The third and fourth are illuminating
for where they go wrong.
Jackendo 's rst objection is that, ``unlike metaphor, thematic paral-
lels [i.e., the spatial thematic relations paralleled in the event and state
elds] are not used for artistic or picturesque eect'' (1983, 209). True; but
Lako et al. (and we) would, of course, deny that metaphor need be used
only for artistic or picturesque eect.
Second, ``thematic structure is the only means available to organize a
semantic eld of events and states coherentlyit is an indispensable ele-
Knowledge by Metaphorical Content 181

ment of everyday thought'' (ibid.). True again (or at least, let's grant the
point). However, the apparent assumption that metaphor is dispensable is
false. As I'll argue in the next section, a classic motivation for the use of
metaphor is catachresis: to ll gaps in the vocabulary of the language for
which there are (at a given time) no (literal) expressions. This may also be
true of whole systems or networks of metaphors that serve to structure
domains that (at the time) possess no structure ``of their own.'' Metaphors
of this kind would be ineliminable and indispensable.
Third, and more interesting, Jackendo argues that Lako et al.
``stretch'' the term ``metaphor'' when it should be limited to expressions
that manifest ``some overt'' or ``literal incongruity,'' be it semantic
anomaly or a pragmatic incongruity (Jackendo and Aaron 1991, 325).
As a test, he proposes that we construct ``diagnostic'' sentences whose rst
clause ``acknowledges the incongruity of the mapping'' and whose second
clause ``constructs a hypothetical invocation of the mapping that moti-
vates the metaphorical reading.'' For example, to capture the metaphor-
icity of
(13) Our relationship is at a dead end,
which would be an instance of the mapping A RELATIONSHIP IS A
JOURNEY, Jackendo proposes the diagnostic sentence:
(14) Of course, relationships are not journeysbut if they were, you
might say ours is at a dead end.
A number of Lako et al.'s ``metaphors,'' he argues, fail this test. The
purported temporal metaphor
?*(15) Christmas is approaching,
based on the mapping TIME IS A MOVING OBJECT, only question-
ably passes the diagnostic test:
(16) Of course, times are not a medium in motionbut if they were,
you might say Christmas is approaching.
The diculty with this argument and diagnostic, as I noted back in
chapter 1, is its presupposition that all metaphors are literally or overtly
incongruous, either pragmatically or semantically. This assumption has
been roundly criticized for the last twenty years, and it is not at all clear
how the diagnostic would apply to sentences that can be both literally and
guratively true (even in the same context), such as (17) and (19). Neither
(18) nor (20) does the job.
182 Chapter 5

(17) The painting is blue.

*(18) Of course, paintings don't have colorsbut if they did, you might
say that the painting is blue.
(19) Jesus is a carpenter.
*(20) Of course, people aren't craftsmenbut if they were, you might
say that Jesus is a carpenter
Finally, Jackendo 's fourth argument is based on the observation that
the most remarkable aspect of metaphor is its variety, the possibility of using
practically any semantic eld as a metaphor for any other. By contrast, thematic
relations disclose the same analogy over and over again. . . . That is, the theory of
thematic relations claims not just that some elds are structured in terms of other
elds, but that all elds have essentially the same structure. (1983, 209)
It is not entirely clear how we should understand this last passage. On
one reading, the claim is that there is one semantic eld (source), the
spatial, in terms of which all, or many, other semantic elds (targets) are
structured. But this fact (if it is one) does not count against that single
source-eld applying metaphorically to all the others; on the contrary,
such a fact would simply testify to its power as a metaphor. Nor would it
exclude the possibility that there are additional mappings (sources) that
might structure those targets, compatible with what Jackendo calls the
``variety'' of metaphor. Nor does the Thematic Relations Hypothesis
deny the possibility of another eld (metaphorically) structuring the spa-
tial. On this reading, the argument is moot against Lako et al.
On a second reading of Jackendo 's objection, the Thematic Relations
Hypothesis is not a claim to the eect that one eld, the spatial, is used to
conceptually structure other elds. Rather it is the claim that there are a
set of ``abstract parameters'' that structure all (or many) semantic elds,
among them (and perhaps more clearly than elsewhere but, in no inter-
esting sense, primarily) language about space. There is, then, no mapping
at work, just one structure realized in dierent elds. Contra Lako et al.,
the existence of parallel thematic relations is no evidence specically for
metaphor.
One problem with this reading is that it leaves unexplained why none-
theless there surfaces so much explicit spatial language in the other
domains, but no or little, say, temporal or possessive language in the
spatial domain. If there were simply one common abstract structure
underlying all these dierent elds, no one of which is privileged over the
others, we would expect much more reciprocal language than we nd:
Knowledge by Metaphorical Content 183

existence or temporal language for the spatial eld, circumstantial talk for
the temporal, or temporal for the circumstantial. The omnipresence of
specically spatial language is signicant.
Despite these diculties, I agree with Jackendo 's conclusion: Lako
et al.'s spatial ``metaphors'' for time, states, and events are not meta-
phorical. But the reason is not that Lako et al.'s mappings do not exhibit
the ``variety'' of metaphor, where that is understood as ``the possibility of
using practically any semantic eld as a metaphor for any other'' (Jack-
endo 1983, 203). The real ``variety'' of metaphor rather consists in its
context-dependence: the fact that any semantic eldthat is, any expres-
sion from one semantic eldcan, in dierent contexts, metaphorically
express dierent contents; or that given (almost) any two semantic elds,
with appropriate context sets of presuppositions, expressions in one can
metaphorically co-occur with the other. What makes it questionable
whether Lako et al.'s spatial mappings are metaphors at present is that
there is no evidence that context plays a role in their present interpreta-
tion. Hence there is no reason to thinkas there ought to be if they were
metaphors (i.e., if they had context-sensitive metaphorical characters)
that if there were dierent presuppositions in their contexts, they would
have dierent interpretations. I emphasize ``at present'' because I do not
wish to make any claim about their history; the claim that they are not
metaphors at present is compatible with the fact that they originally
entered the language as metaphors. On the other hand, I am also not
objecting (as many others have objected in reaction to Lako et al.) that
Lako et al.'s conventional metaphors are at present nothing more than
dead metaphors. By denying that they are metaphors, I do not wish to
imply that as a system they are in any way ``dead.'' The mappings are
very much alive in that, as both Jackendo and Lako et al. demonstrate,
they actively interact with syntax, and hence with the language faculty, as
well with our other belief systems and action. Yet, although the mappings
are alive, they are not metaphoricalbecause they exhibit no (propensity
to) context-dependence. For this reason, Lako et al.'s mappings are too
broad to capture the specically metaphorical.
Lako et al.'s mappings are also too general. They fail to account for
the specicity of dierent metaphorical interpretationsand, again, pre-
cisely because they fail to take into account the role of the context. To
illustrate the problem, let's take a closer look at Lako et al.'s analysis of
a metaphor discussed at length by Glucksberg and Keysar (1990):
(21) My job is a jail.
184 Chapter 5

G&K analyze this metaphor as a ``class-inclusion'' statement, or predica-

tion, in which the subject referred to by `my job' is assigned to a category
referred to by `a jail'. Now, psychologists like G&K speak of categories
where we speak of properties or conceptslargely as a function of our
dierent theoretical interests: theirs with the processes governing catego-
rization, ours with the content of utterances. But the idea is the same: The
phrase `a jail' is used here to refer, not to its ordinary extension ( jails) but
to a superordinate category (or property) of which that extension is an
exemplary memberin our terminology, the predicate expresses a prop-
erty (presupposed to be) exemplied by its (literal) extension.47
But how do we know which superordinate category or property is re-
ferred to by `a jail'? Although G&K do not mention exemplication or
sampling, they note correctly that the category referred to depends on the
other objects with which the (literal) object is classied and on the other
categories, or alternatives, with which the given category-expression is
grouped. For example, (1) a jail can be (more or less literally) a spartan,
unluxurious dwelling, in contrast to palaces, hospitals, resorts, dormi-
tories, and hotels. Or (2) a jail can be a kind of punishment, classed along
with scoldings, spankings, trac tickets, and capital executions. This is
what `jail' means when my ve-year-old son, who has hit his brother, an-
grily blurts out when I tell him to go to his room:
(22) OK, send me to jail!
Or (3) a jail (or prison) can contrast with open roads, a life in which
people can freely come and go as they choose. Thus a Palestinian ocial
once described his town of Bethlehem as a prison, because he could not
come and go without rarely approved Israeli permits. Or, nally, (4) a jail
can be classied among psychologically pleasant, neutral, and unpleasant
situations. For lack of a better name, `a jail' in this last scheme of classi-
cation would refer to situations that hold individuals against their will,
that oer no satisfaction, that the individual would leave if he only could
(but can't), that are psychologically conning. The contrasting elements
in this schema include, perhaps, honeymoons, holidays, and vacations.
This also may be the sense of `prison' when Hamlet announces
(23) Denmark's a prison. (Hamlet II, ii, 260)
As these examples illustrate, the specic superordinate category to which
a given expression refers depends on the schema of alternatives with
which it alternates. If we think of the exemplication-schema for an ob-
ject (or for its term) as its context, this is one way in which this kind of
reference is context-dependent. But, in a broader sense, the classication
Knowledge by Metaphorical Content 185

of `a jail' with one or the other of these exemplication-schemas also

depends on contexton other features of the discourse, identities of the
participants, etc.
Thus in dierent contexts, depending on what we presuppose about the
job, speaker, and discourse, we will assign `a jail' in (21) to dierent
schemas relative to which it exemplies dierent properties and meta-
phorically expresses one or the other of the following contents: Either that
my job is a punishment with a term (say, for having committed a crime-
like act like dropping out of school as a youth, thereby committing me to
the consequences of my error); or that my job involves a spartan exis-
tence, with few amenities, low pay, no frills or perks; or that the job is
claustrophobic, heavily constrained, with little opportunity for spontane-
ity; or, last of all, that it is simply an unpleasant situation in which I am
trapped. In short, exemplication metaphors like (21) are highly context
sensitive, expressing dierent contents given dierent presuppositions
concerning the exemplication-schema involved.
Now, Lako (1993, 236) claims that he can neatly explain the same
metaphor `my job is a jail' in terms of the interaction of three conceptual
metaphors: ACTIONS ARE SELF-PROPELLED MOVEMENTS (an
instance of the event-structure metaphor), PSYCHOLOGICAL FORCE
IS PHYSICAL FORCE, and GENERIC IS SPECIFIC. His reconstruc-
tion of the interpretation falls into ve steps:
1. We begin with our ``knowledge schema,'' that is, our community-wide
presuppositions about jails that includes ``knowledge that a jail imposes
extreme physical constraints on a prisoner's movements'' (ibid.).
2. The GENERIC IS SPECIFIC metaphor next applies, ``factoring out''
what is specic to prisoners and jails, thereby yielding: X imposes extreme
physical constraints on Y 's movements.
3. The ACTIONS ARE SELF-PROPELLED MOVEMENTS metaphor
next applies to the result of 2 to yield: X imposes extreme physical con-
straints on Y 's actions.
4. The PSYCHOLOGICAL FORCE IS PHYSICAL FORCE metaphor
next applies to the result of 3 to yield: X imposes extreme psychological
constraints on Y 's actions.
5. Substituting `my job' for X and me for Y yields as an interpretation for
`my job is a jail': ``my job imposes extreme psychological constraints on
my actions.''
This analysis raises a number of questions, beginning with the rst and
second stages. First, GENERIC IS SPECIFIC may be a ``mapping,'' and
even a mapping involved in the interpretation of some metaphors, but it is
186 Chapter 5

not itself a metaphor. If the mapping were a metaphor, it ought to be

realized or realizable linguistically in a metaphorical expression. What
would it be? Lako 's own example of a generic metaphor is, say, `death is
destruction'. Analogously, we might propose: `A situation that imposes
extreme physical constraints on someone's movements is a jail'. A bit
odd, yes; but a metaphor? Hardly. The problem lies in the origins of
Lako 's idea of generic structure, which he rst introduces as a constraint
on mappings: as a set of generic-level conditions (overall event shape or
causal and aspectual structure) roughly like thematic relations.48 Only
later, when he proposes a ``mechanism'' corresponding to ``our ability to
extract generic-level structure,'' and attempts to assimilate all such mech-
anisms to his rubric of mappings, does he invent the GENERIC IS SPE-
CIFIC ``metaphor.'' This is unnecessary. We already have the needed
``mechanism'' in the (not-language-specic) relation of exemplication,
which in the case of events is, as we all agree, constrained by thematic or
generic-level structure.49 Hence there is no evidence that the GENERIC
IS SPECIFIC mapping is itself a metaphor.50 And insofar as Lako 's
own account that rests on mappings like GENERIC IS SPECIFIC also
relies, as we just said, on a reference-relation like exemplication (whether
or not he calls it by that name), it is no genuine alternative to G&K.51
The main and least controversially metaphorical mapping in Lako 's
analysis is PSYCHOLOGICAL FORCE IS PHYSICAL FORCE. But
this metaphor enters into the story only because Lako begins by assum-
ing that our ``knowledge schema'' about jails specically includes only the
``knowledge that a jail imposes extreme physical constraints on a pris-
oner's movements'' (ibid., my italics). Why couldn't the ur-schema equally
well include the knowledge that a jail imposes extreme psychological
constraints (social isolation, limited access to family and outsiders) on a
prisoner? Once we include that information in our initial knowledge-
schema there is no work left to be done by the PSYCHOLOGICAL
FORCE IS PHYSICAL FORCE metaphor.52
Finally, let's assume that Lako et al.'s interpretation for `my job is a
jail'namely, ``my job imposes extreme psychological constraints on my
actions''is in fact one of its metaphorical interpretations. Still it is only
one among a numberdepending on the exemplication schema in which
jails are taken to function in the context of utterance. As we have just
seen, dierent schemata yield rather dierent metaphorical interpretations
for the one sentence. Compared to these riches of alternative interpre-
tations, Lako et al.'s interpretation is much too thin, monolithic, and
one-dimensional.53 The reason, as I hinted earlier when I said that their
Knowledge by Metaphorical Content 187

linguistic metaphors are expression types rather than interpretations of

tokens, is precisely that their theory of conventional, conceptual metaphor
fails to do justice to the role of the context in metaphorical interpretation.
In conclusion, Lako et al.'s theory of conceptual metaphors draws
from a rich body of evidence about metaphorical networks. But it is far
from evident that the ``mappings'' to which the theory appeals are spe-
cically metaphorical, that the principles that govern its mappings are
really dierent from those of opposing accounts like G&K's, and that it
does justice to the role of context, which is necessary to account for the
full variegated range of metaphorical interpretations. On the other hand,
those who characterize the dierence between Lako et al.'s theory and
G&K's (or mine) as a dispute over whether metaphor ``involves the con-
struction of novel categories'' or whether it ``always relies on preexisting
conceptual mappings'' are, I think, also o the mark.54 Just as Lako
et al.'s theory must be supplemented by (or already involves) context-
sensitive factors like exemplication, so the notion of exemplication it-
self relies on and interacts with networks like Lako et al.'s. Some of
these networks, such as exemplication-schemas, may be specic and
novel to the context of the metaphorical utterance. Others, like the the-
matic relations of the metaphor and its inductive network, may preexist
the particular context, either because they are built into our language
faculty or because they are based on established communally shared
associations. In either case, the dierent networks, and dierent accounts
that focus on dierent networks, complement rather than compete with
each other.

VI Exemplication, Catachresis, and De Re Knowledge by Metaphor

With the analysis of exemplication in hand, I can now draw out an

analogy between metaphors and predicate demonstratives to which I
alluded in chapter 3. In the course of our discussion of the direct reference
thesis for demonstratives, I pointed out that Kaplan's characterization of
a singular proposition, as a proposition that contains at least one indi-
vidual in contrast to a general proposition that contains only properties
(or universals) and logical functions, fails to do justice to sentences con-
taining natural kind or substance terms or predicate demonstratives like
`is that F' (e.g., `the table in Classics 12 is that shape[points at a long
rectangular board]' or `thus' (e.g., `he broke the board thus[snaps a thin
twig between his two ngers]'). On the one hand, the contents of these
expressions are properties (kinds, substances, other abstract entities)the
188 Chapter 5

stu of general propositions. On the other, these expressions appear to

express (or refer to) their respective properties just as directly referential
terms refer to their individual contents: without the mediation of a purely
conceptualized representation whose conditions are satised, the kind of
relation that characterizes indirect reference. In this respect, the contents
of predicate demonstratives are more like the constituents of singular
propositions.
To address this problem, I proposed a broader distinction between ref-
erential and purely conceptualized propositions (under which we would
subsume the singular/general distinction). Following Frege, we assign all
expressions referents, the referents of n-ary predicates being n-ary prop-
erties. The class of referential propositions now includes all singular
propositions, but it also includes all propositions at least one of whose
constituents is a property and relation ``in itself ''a bare property or re-
lationrather than a conceptualized representation whose qualitative
individuating conditions in turn determine the property or relation via
satisfaction at a circumstance. This distinction is more epistemological
than Russell's, which distinguishes the two classes of propositions solely
by the identities of their constituents. Ours also takes the dierent classes
of propositions to mark dierences between the means by which we con-
ceive of their constituents. As I said in chapter 3, the distinction might
reect a number of dierent epistemic situations. Sometimes we possess a
purely qualitative conceptualization of the property that enables us to
denote it. At other times, as with many directly referential (singular)
terms, it is only by employing our contextual relations to the property
that we are able to express a proposition containing it. We may be able to
express the property only by way of a sample that we demonstrate in
context or by way of applying the property to a particular object in a
context, without knowing, and perhaps even without there being in the
language, a label or concept for that property. In these cases, to borrow a
usage of Tyler Burge (1977), we refer de re to the property. On the other
hand, we refer to the property de dicto where we express or refer to it by
using a purely conceptualized individuating representation, where we ex-
press it by using vocabulary in our linguistic, or conceptual, repertoire
that is determinately true or false of arbitrary things without depending
on features of its context of application. This dierence between de re and
de dicto expression of or reference to properties is a matter of how they
are expressed or referred to in the utterance and by its agent. The same
property expressed de dicto by one sentence on one occasion may be
expressed de re by another. De re expression via the context may also
Knowledge by Metaphorical Content 189

have other virtues not due (simply) to the lack of de dicto conceptual or
linguistic resources. Context-sensitive character, as I'll show in chapter 7,
may also carry its own cognitive signicance in addition to its proposi-
tional, truth-conditional content (in a context). A speaker may there-
fore choose to express a given property de re, using a context-sensitive
expression, in order to communicate this additional character-istic
information.
Some utterances containing metaphors also express referential propo-
sitions. Like predicate demonstratives, they express or refer to their con-
tents by exploiting (via their nonconstant characters) features of their
nonlinguistic contexts, without the mediation of purely conceptualized
representations denoting their respective contents (in extension) at each
circumstance of evaluation. As linguists have often observed, one impor-
tant function of metaphor is catachresis:55 to ll in lexical gaps or to
compensate for the lack of vocabulary in a language for a concept or
property. The deciency is not simply a shortage of words. Rather it sig-
nals the lack of representations in our conceptual or linguistic repertoire
whose characters are characters of those properties, characters that would
determinately express (or refer to) those properties in all contexts, inde-
pendently of their application to a particular thing in a particular context.
Instead the metaphorical expression expresses the property by exploiting
features specic (and sometimes unique) to its context. In particular, the
metaphor may employ contextual presuppositions about what property
(whether or not we know a name for it) is sampled or exemplied by
things in the context. Given my account of exemplication, we can ll in
the details of this picture more fully.
Let me begin with a supercially similar example of nonmetaphor,
which Glucksberg and Keysar (1990, 1993) compare to the mechanism by
which metaphor enables us to express, or refer to, novel superordinate
categories (or properties). The dierence between the two cases is
instructive.
G&K (1990, 1993), following Rosch (1973), point out that all lan-
guages have simple names for basic level objects (such as `table', `chair',
`bed') although some languages such as American Sign Language lack (or
did, until recently) names for superordinate categories (such as `furni-
ture'). To refer to the superordinate category of furniture, ASL signers use
compounds or concatenations of the ``basic objects signs that are proto-
typical of that category'' (1993, 409) such as `table-chair-bed'. So, to say
`I lost all my furniture in the house re but one thing was left: the bed',
ASL signers produce the string:
190 Chapter 5

House re [] lose all chair-table-bed etc., but one left, bed. (Newport and Bellugi
1978, 62)
ASL, then, contains a mechanism to compensate for its gaps in lexical
vocabulary, by using concatenations of names of prototypical or exem-
plary basic level objects of a category as a name for the category itself.
Similarly, G&K claim, metaphor serves as a compensating mechanism in
natural language. Lacking a name in English for a novel superordinate
category, we use the name of a prototypical or exemplary member of the
category (e.g., `is a Kennedy' or `is a walking time bomb') to (metaphor-
ically) name the category itself.
There is a dierence, however, between the two cases. In ASL the lack
of simple superordinate category names is purely lexical. This lexical de-
ciency is not because users of ASL do not understand the concept of
furniture, not because they lack knowledge of the conditions under which
something belongs to the superordinate category. It may be vague
whether something borderline is furniture because it is vague whether it is
a bed or vague whether it is a chair or vague whether it is a table, etc. But
the vagueness of the superordinate category of furniture is neither more
nor less than the vagueness of the basic level object categoriesbed,
chair, table. There is no more deciency in the ASL user's understanding
of the superordinate category `furniture' than there is in his understanding
of the basic level categories. It is, to be sure, striking in the ASL case that
the lexical deciency is not simply of one particular piece of vocabulary
but of names of categories of a whole type or level, namely, the super-
ordinate. Nonetheless the catachresis is simply lexical.
Metaphor is a dierent story. With metaphors of exemplication, we
also use an expression that (literally) refers to one thing to express or refer
to a property that (it is contextually presupposed) the referent exemplies
in the absence of a (simple) predicate that (literally) expresses the prop-
erty. But in at least a central set of cases of this sort, the property has no
linguistic expression to call its own, not simply for lack of a word but for
lack of adequate understanding. This is especially true when the metaphor
is used to express or refer to a ``novel'' property. What makes the prop-
erty novel is not that it is being applied for the rst time to a particular
object but that we can apply it only in relation to the particular object
that samples, exemplies, or displays it in context. Such a property is not
yet fully conceptualized. We lack at the time of utterance a concept, or
representation, of the property that would enable us to determine what
possesses it in a context-invariant or transcendent manner. We lack
knowledge of the conditions that would enable us to determine whether
Knowledge by Metaphorical Content 191

an arbitrary object possesses the property regardless of the extralinguistic

context in which we make the determination. We can only express or refer
to, say, a color as the color of an object to which we demonstratively refer
or gesture. Like a complete demonstrative that enables us to refer to an
individual by employing nonconceptual nonlinguistic features, the char-
acter of the metaphor, given a complementary set of presuppositions,
enables us to ``point'' to the property even when we don't possess the fully
conceptualized linguistic means to express it independent of context.
What would it be to have such a fully conceptualizedin one sense of
the word: literal (as I'll argue in ch. 8)means to express a property? I
don't have a good answer to this question. But to return to our point of
contrast with ASL, it is not merely a matter of having a (literally inter-
preted) word for the property, although the availability of and ability to
apply such a context-independent word is frequently evidence of having
such an epistemic capacity. It is therefore misleading to think of the de re
use of metaphors of exemplication as nothing more than verbal cata-
chresis, a matter of plugging a hole in the literal vocabulary. Instead the
capacity is a function of the extent to which our knowledge of the con-
cept, or conception, of the property transcends any particular application
of the predicate that expresses it. Among other things, this capacity is re-
lated to the degree to which the concept of the property is integrated
within our conceptual repertoire, roughly, the degree to which we know
what the concept entails and what entails it among a relatively rich num-
ber of logical relations to other concepts and beliefs. This is, to be sure, a
very rough statement of the idea of fully conceptualized understanding,
but where metaphors serve to introduce properties to which we are related
only de renot fully conceptualized propertiesit is important that we
have some sense of what would be required to transform the metaphori-
cally expressed property into one that is (more) fully conceptualized.56
Let me illustrate the epistemic dierence between the de re and the de
dicto with another example. Typical novel metaphors express not just one
property but, as G&K (ibid., 421) note, ``a patterned complex of prop-
erties in one chunk,'' all those the literal referent of the metaphor ex-
emplies. The, or a frequent, way the metaphor expresses that complex
property, or complex of properties, is not by expressing the totality of
constituents as one conjunction or articulated compound but simply as
the undierentiated chunk of properties exemplied by Fs. The distinc-
tion between these two modes of expression parallels Gareth Evans's
(1982, 284289) distinction between two types of memory, recall and
recognition. In some tasks, when asked to remember or identify Fs, sub-
192 Chapter 5

jects have the capacity to recalland list seriallydescriptive features of

Fs individually and without prompting, features the satisfaction of which
would make something a F. In other tasks, however, subjects cannot re-
call the individual features, but they can recognize Fs on presentation.
Evans argues that memory and identication need not require recall, that
recognition is frequently sucient, and, furthermore, that even where
subjects can recall the features of something, they often identify it by
simply recognizing it without employing knowledge of the features they
can recall.
Similarly with metaphors of exemplication: Speakers and interpreters
can often recognize the property-complex metaphorically expressed by
`Mthat[F]' in a context cthe property exemplied by Fs in cwithout
being able to recall, or articulate, the constituents of the property-com-
plex. Ted Cohen (1990) describes how his father could immediately and
repeatedly recognize whether a beer was green. But he could never tell you
why, or state in other terms what that metaphorically expressed property
was. This was not because he was just guessing, nor a matter of a single
occurrence of the metaphorical predication, nor something idiosyncratic
or necessarily subjective about the metaphor. There was a common taste
or feel to the beer, a taste or feel he could recognize on dierent occasions
and that he could even induce others to recognize. But he (and others)
didn't have any way to articulate that metaphorexplicate the features
that make a beer greenexcept by applying it to particular beers in con-
text. One function of metaphors is just this: to refer to categories (or
properties) for which we do not at the time have the capacity to articulate
explicitly the conditions in virtue of which things belong to (or have) it.57
The interpretation of such metaphors involves a kind of de re grasp of
their contents, recognition that relies on and is sensitive to the cues
aorded in the context, the ``lighting'' and ``angle'' in and from which the
property is presented.
With this idea of de re knowledge by metaphor, we can also understand
better one well-known applied problem of metaphor: ``irreducible'' meta-
phors in theological discourse.58
The problem of irreducible metaphors in theological language arises
from a combination of assumptions. First, there is the assumption that ``it
is possible to make true statements about God, . . . to convey in words
some apprehension, however inadequate, of what God is like.''59 Second,
it is assumed that it is impossible to say anything true about God
expressed by language used literally, that is, language whose conceptual
content is fully determined by the linguistic meanings of the words used.
Knowledge by Metaphorical Content 193

Given these two premises, it is next proposed that oneperhaps the

onlyway to make true statements about God is by metaphor. But given
the second assumption, it is further inferred that ``what is said in the
metaphorical utterance cannot be said, even in part, in literal terms''
(ibid.). Hence, if there are true metaphorical statements about God, they
must be irreducible. Their content cannot be said, even in part, in literal
language.
The ``problem'' of irreducible theological metaphors is that this con-
clusion, it is next argued, is indefensible. ``What is said'' by the metaphor
that is claimed to be irreducible is also claimed to be its propositional
content, its truth-valued content. (Recall that, according to the rst
assumption, the metaphorical statement is indeed true.) But if the theo-
logical statement, say, `God is a rock', has propositional content, then
corresponding to the (metaphorically used) predicate, there must be a
property P that is attributed to God. But if there is such a property P,
there should also be, at least in principle, a predicate that can be literally
used to attribute P. That is, there may or may not always be available to
the speaker such a predicate expressing P (depending on a variety of lin-
guistic and extralinguistic factors), but a speaker who attributes P to God
must at least have a concept, or conceptual representation, of Pand
once she has such a concept, it always ought to be possible at least in
principle to associate it with a predicate as its meaning, that is, with a
literally used predicate. Therefore, there could not be a property expressed
by a metaphor that cannot be expressed at all, even in principle, by a lit-
erally used expression. Hence there are no irreducible metaphors.
Another way of putting the argument would be this: Either the meta-
phor expresses a truth-valued proposition or it doesn't. If it doesn't, it is
irreducible only in the trivial sense that it has no content to express liter-
ally. If it does, then it must be possible to state that content, like that of
any proposition, in literal languagefor that is the very nature of prop-
ositional content. Either way there can be no irreducible metaphors about
God. The theologian cannot have his metaphorical cake and eat it, too.60
The fallacy in the argument is, of course, its assumption that all prop-
ositional content must be fully conceptualized, if it is conceptual at all. In
referential propositions, the constituent corresponding to a metaphor may
be a bare property for which the speaker possesses no fully conceptualized
representation. Nonetheless there is denite reference to, or expression of,
the property, and there should be a fact of the matter whether that prop-
erty is true of God, even though we may not know whether it is. Indeed, in
light of our epistemic gloss on de re attitudes, divine properties would be
194 Chapter 5

exactly the sorts of things we ought to expect to be expressible (only) de

re. Theologians (or other believers) who want to attribute properties to
God do so on the basis of various kinds of arguments. But given the
transcendence of the deity, they must also acknowledge that, despite their
arguments, we can't really fully understand the nature of those divine
properties. Propositions about divine properties are, then, paradigms of
beliefs or thoughts that we incompletely understand (or even completely
fail to understand), that is, of not fully conceptualized beliefs or thoughts.
Nonetheless, through the context-sensitive mechanisms of metaphor, we
can ``point'' to those properties despite our conceptual deciencies.61
Let me add one more loop to this story to drive home the point that it is
identiable properties that are the constituents of these metaphorically
expressed de re propositions, that is, properties that, despite our lack of
concepts, theologians appear to believe we can identify or reidentify (if
only in context) on dierent occasions.62 Although divine properties are
properties true of God, not all metaphorically expressed propositions
containing divine properties need be propositions about God. Consider
the biblical verse, ``There will you worship gods made by human hands
out of wood and stone, gods that neither see nor hear, neither eat nor
smell'' (Deut. 4, 29). Although this verse is true under its literal interpre-
tation, the great medieval Jewish Talmudist, mystic, and biblical com-
mentator R. Moses Nahmanides argued that what are denied of the false
gods in this versethat is, the properties expressed by `seeing', `hearing',
and `eating'are the properties expressed by these predicates when they
are interpreted metaphorically as applied to God, say, in a sentence like
`God sees' or `God eats [your sacricial oerings]'. For Nahmanides,
then, we can meaningfully speak of the same property truly attributed to
God and denied of false gods. But such a divine property is not purely
conceptualized, not fully understood, and it is assumed that there is no
literal expression for it. To generate the proposition, we need, then, a
context of interpretation in which we hold certain presuppositions about
God, even though the (negated atomic) proposition itself predicates the
divine property of something other than God.
In concluding this chapter, let me add two nal comments: First, the
not fully conceptualized properties that a speaker can express or refer to
de re using metaphors of exemplication constitute only one kind of
knowledgepropositional contentexpressed by metaphor. Not all
metaphors are of this de re type, and this is also not the only kind of
knowledge conveyed by metaphor. Many metaphors express properties
for which we do have fully conceptualized representations. And even
Knowledge by Metaphorical Content 195

some metaphors of exemplication express or refer to contents we could

express de dicto, using a fully conceptualized representation. In some of
these cases, as well as with de re metaphors of exemplication, the char-
acter of the metaphor carries a second kind of knowledge by metaphor.
This kind of cognitive signicance is additional to its propositional con-
tent in a contextor so I'll argue in chapter 7.
Second, I have emphasized how exemplication can reveal, or render
cognitively accessible, bare properties that we have no fully concep-
tualized means to express. But exemplication (and metaphor) can also
serve to conceal a fully conceptualized and independently understood
property. This second function of metaphor, concealment, is hardly men-
tioned nowadays, but it was a classical and widely cited explanation for
the use of metaphor in Scripture and other sacred writings by medieval
philosophers, especially those in the Islamic and Jewish Aristotelian tra-
ditions. Moses Maimonides, for example, explains that ``Plato and his
predecessors . . . concealed what they said [from the multitude] about the
rst principles and presented it in riddles. . . . [E]ven those upon whom the
charge of corruption would not be laid in the event of clear exposition
used terms guratively and resorted to teaching in similes'' (1963, 4243).
In these traditions, then, metaphors and other gures are primarily
usedand note, by no less than philosophersin order to conceal philo-
sophical truths from the community at large, both to protect the
uneducated from knowledge for which they are unprepared (on the as-
sumption that a little knowledge can harm) and to protect themselves
from charges of corruption leveled by the uneducated. In a slightly dif-
ferent vein, Pascal also writes that
God being unwilling to reveal these things to these people who were unworthy of
them, and yet wishing to eect them so that they should be believed, foretold the
time clearly and at times expressed them clearly, but very often in a gurative way,
so that those who loved the symbols should go no further and those who loved
what was symbolized should see it.63
For Pascal also, metaphors simultaneously conceal from some, the un-
worthy, and reveal to others, believers. The latter, who know and seek the
content of the metaphor, ``see'' it in the gure. Knowing the intended
content and that the expression is a gure, they recover the context that
must obtain in order to interpret the gure that expresses the content. The
unworthy, however, simply love the symbol itself, perhaps without even
recognizing that it is a metaphor. Being unworthy, they cannot grasp its
content.
196 Chapter 5

Here, again, a comparison with demonstratives is helpful. Suppose I

have a delicious piece of gossip about Phil to tell a colleague at a crowded
party but I am afraid that the wrong ears may pick it up. Instead of using
Phil's proper name, I use the complex demonstrative `that abusive liar'.
The demonstrative both enables me to communicate the identity of my
intended referent to any listener who shares my context (i.e., pre-
suppositions) and conceals that identity from everyone else outside that
context. Likewise metaphors of exemplication. Knowing the property
exemplied on an occasion depends on knowing its context: the relevant
sample schema and the range of features the schema sorts. A speaker may
therefore choose to use a metaphor of exemplication in order to restrict
the audience to those who can identify its context. Were she to express
that property with a literal expression, she would not be able to exercise
the same control over its dissemination. In sum: The specically contex-
tual orientation of a metaphor enables it both to extend our powers of
expression and to restrict the reception of their contents. That these two
complement each other may be no accident.
Chapter 6
Metaphorical Character and
Metaphorical Meaning

In the course of my critique in chapter 2 of Donald Davidson's use theory

of meaning, I formulated a number of semantic problems involving
metaphors in order to motivate the turn to metaphorical meaning. If, as I
argued, one task of meaning is to (semantically) constrain the contents of
possible interpretations, and we can demonstrate the need for semantic
constraints on possible metaphorical interpretations, that would be an
argument for the existence of metaphorical meaning. With our semantic
theory in hand, we are now in a position to say more precisely how these
constraints work and, thus, what we know when we know the meaning of
a metaphor. We know the metaphorical character of the expression, that
is, the character of the expression used, or interpreted, metaphorically (or,
equivalently, the character of its underlying metaphorical expression).
Like the character of an indexical, the character of the metaphor con-
strains its content by limiting its contextual source to actual presupposi-
tionseven when we evaluate its truth in nonactual circumstances. As
with the character of an indexical, there is also knowledge conveyed spe-
cically at the level of metaphorical character, information or signicance
in addition to that conveyed by the (propositional) content of the meta-
phor in a context. In this chapter, I'll discuss the rst of these roles of
character qua meaning, its role in our knowledge of metaphor. In chapter
7, I'll address the second role, its part in our knowledge by metaphor.
I begin by explaining the kind of meaning with which the `Mthat'
operator equips us, by analogy to Kaplan's `Dthat'. I next turn to the
dierent ways we can fail to ``get'' a metaphor in order to sharpen our
focus on what we know when we know the character of a metaphor. Of
these various kinds of metaphorical incompetence, lack of knowledge of
character turns out to be a failure of understanding best described as not
knowing the meaning, or semantic incompetence. Using the metaphorical
198 Chapter 6

character-content distinction, I then show how we can explain the data

and solve the problems introduced in chapter 2. In contrast to those who
take metaphor itself to be a kind of speech act, I also argue that an inde-
pendent notion of metaphorical meaning is necessary precisely to account
for (indirect) speech acts performed with metaphors. Finally, I'll turn to
metaphors whose character does not seem to t our model (e.g., nomina-
tive metaphors like `the sun', referring to Juliet, in `the sun is ascending')
and to tropes other than metaphor (such as simile and irony) to raise the
question of whether our semantics ought to be extended to them. Put
dierently, this is the question of whether there is one natural kind that
subsumes all tropes and whether all gurative interpretations should be
explained in one way.

I Knowledge of `Mthat'

Suppose a speaker S knows the metaphorical interpretation of one ex-

pression F but no othersand suppose that the reason is not that S
doesn't know the relevant m-associated presuppositions for the others.
Even if S's interpretation of F is an interpretation that would unques-
tionably count as metaphorical were it produced by a speaker with meta-
phorical competence (e.g., someone who can freely interpret expressions
metaphorically), S's own interpretation of F is not metaphorical. Meta-
phorical competence involves mastery of a general skill that one can
apply to arbitrary expressions across the language. More theoretically,
metaphorical competence consists in knowledge of a schematic rule that
applies to all expressions that admit a metaphorical interpretation.1 This
schematic rule governs the characters of all metaphorical expressions of
the type `Mthat[F]', for each substitution-instance of F. The metaphori-
cally competent speaker knows how to generate metaphorical expressions
given her knowledge of the expressions F to be interpreted metaphori-
cally. So, if F is an expression of a syntactic category C (e.g., a one-place
predicate) with an interpretation of type T (e.g., a property), she knows
that `Mthat[F]' will belong to the same syntactic category and have an
interpretation of the same type.2 Either one knows this schematic rule and
has the general skill or one doesn't. It makes no sense to say that one
could know this rule, or have this skill, for some expressions but not for
others. Thus part of what the speaker knows when she knows how to in-
terpret a metaphor F is not simply something about the single expression
F. Her knowledge of metaphor is closer to knowledge of an operator, of
an interpretive operation she can perform on any (literally interpreted)
Metaphorical Character and Metaphorical Meaning 199

expression. And in this respect the character of a metaphor is more like

that of the demonstrative interpretation operator `Dthat' than of that of
any individual indexical (type) such as `I' or `here'.
If we recall our earlier discussion of `Dthat', however, other issues may
prima facie appear to complicate the question whether `Mthat' is an
operator. Both dthat-descriptions and metaphorical expressions, we said,
are attempts to construct lexical representations corresponding to, or en-
coding, uses (or interpretations) of language. In chapter 3, I mentioned
various diculties that arise with dthat-descriptions stemming from this
shift from use or interpretation to lexical representation. On the one
hand, dthat-descriptions are meant to capture directly referential interpre-
tations of the descriptions, on the formal model of directly referential
complete demonstratives (i.e., pure demonstratives like `that' completed
by extralinguistic demonstrations). That, in turn, would suggest that
dthat-descriptions are syntactically and semantically simple expressions
like the individuals who are their simple propositional contentsand I
mentioned two ways to cash out this suggestion. Either treat the complete
dthat-description as syntactically and semantically (though perhaps not
lexically) unstructured, or let `Dthat', like `That', be the only syntactic
and semantic unit and the embedded description, simply a pragmatic
``aside,'' in Kaplan's words, a parenthetic stage direction.
On the other hand (as I argued in ch. 3), there are various reasons to
structure the character of a dthat-description as a complex built up from
the character of the constituent description. First, the dthat-description
is formally constructed (in Kaplan's ``Logic of Demonstratives'') as an
operator taking the denite description as operand. Second, only if we can
recover the character of the embedded description as a separable semantic
entity can we account for the cognitive value of the dthat-description (and
thereby solve Frege's puzzle). Third, because dthat-descriptions are
directly referential but not thoroughly nondenotational, we need to ac-
knowledge the description as a signicant constituent in order to explain
how the direct referent of the dthat-description is xed in (the circum-
stance of ) its context as a function of the denotation of the description.
All these reasons argue that dthat-descriptions ought to be articulated as
syntactically and semantically complex expressions.
There are, then, conicting properties of and motivations for the dthat-
description that pull us in opposite directions. However, rather than con-
demn the dthat-description itself as an incoherent attempt to represent
lexically how we use, or interpret, nondemonstrative expressions demon-
stratively, we concluded in chapter 3 that (like the notion of quotation
200 Chapter 6

that is beset with similar tensions) the fault more likely lies in our under-
standing, and theory, than in the construction itself.
Some of these diculties plaguing dthat-descriptions do not carry over
to metaphorical expressions. Unlike dthat-descriptions, metaphorical
expressions are thoroughly nondenotational; like indexicals, they are
parametric. The contextual feature that the character of the meta-
phorical expression `Mthat[F]' maps into its contentthe context set of
m-associated presupposed propertiesis not something its constituent F
denotes; neither are the presuppositions directly related to the object(s)
denoted by the expression F in c, or its extension. If they were, we ought
to be able to substitute a co-denoting or co-extensive expression for F
without aecting the truth-value of the metaphorical interpretation; but
as we saw from the failure of substitutivity arguments back in chapter 2,
that is not the case. Instead, like the parametric values of the indexicals,
the content of a metaphor is assigned or mapped from a given kind of
contextual parameter, namely, the presuppositions m-associated with the
word F (in c) or, more precisely, the presuppositions associated with the
character of F (in c). This assignment, like that of its value to a variable,
involves no satisfaction or tting of conditions spelled out in the character
or content of the metaphor. But, unlike variables, it involves semantic
knowledge. As with the indexicals (e.g., `now', whose relevant parameter,
the time of the context, is known by the speaker as part of his semantic
knowledge), the relevant parameter (namely, a context set of presup-
positions) is semantically specied, known by the interpreter as part of his
knowledge of language. Nonetheless, because metaphorical expressions
are parametric and thoroughly nondenotational, we don't have the same
compositional motivation to acknowledge an autonomous semantic con-
tribution by the embedded expression (via denotation) to the semantic
value of the metaphorical expression. In this respect, there isn't the
same motivation to separate o `Mthat' as an operator from the em-
bedded expression as operand. Although `Mthat' lexically generates
Mthat-expressions, we can treat them as syntactically simple.
On the other hand, it should be kept in mind that the character of each
metaphorical expression of the form `Mthat[F]' that results from sub-
stituting an actual expression for F is individuated by the character of
that constituent expression; hence each distinct expression (type) inter-
preted metaphorically will yield a metaphorical expression with a distinct
character. For example, the character of `Mthat[`is the sun']' is dierent
from that of `Mthat[`is the moon']'. Therefore, our semantic knowledge
of metaphorthe semantic knowledge represented by the schematic rule
Metaphorical Character and Metaphorical Meaning 201

for `Mthat[F]' that underlies our ability to interpret all expressions meta-
phoricallyis the same for all expressions we interpret metaphorically.
But because each particular metaphorical expression that is a substitu-
tion-instance of `Mthat[F]' has its own respective character, the knowl-
edge by metaphor that is conveyed, and individuated, by the character of
each such metaphorical expression will be distinct to it.
Despite its dierences from `Mthat', `Dthat' furnishes us with a valu-
able model of how we can represent context-dependent interpretations (or
uses) of classes of expressions that are not otherwise indexical or demon-
strative. This is the primary point of the parallel with `Mthat'. So long as
we keep in mind that the content of the metaphorical expression is not
determined compositionally from the contents of its constituents, there
is no reason not to think of a metaphorical expression, like a dthat-
description, as a complex expression formed by an operation on the
constituent expression, that is, as an expression whose character is indi-
viduated by the character of its embedded expression.

II Metaphorical Incompetence3

One way to articulate the kind of meaning an interpreter grasps when he

knows the character of a metaphor is to identify what he fails to under-
stand when he lacks competence in metaphorical character. We can locate
that kind of incompetence among four basic ways in which a speaker/
interpreter can fail to ``get'' a metaphor, four species of metaphorical in-
competence and four correlative sources of possible disagreement among
interpreters over claims expressed metaphorically. The fourth of these is
best described as a failure to know the meaning of the metaphor.
1. Cases where we fail to recognize or identify an utterance u of a sen-
tence s as containing an expression F that is to be interpreted metaphor-
ically. This is a general failure to know-that-metaphor (ch. 1, sec. I), but
we might distinguish three special subcases:
A. I hear u but fail to recognize that it can be interpreted metaphorically,
that it has a (grammatically) possible interpretation that contains at least
one metaphorical interpretation of a constituent. Such a failure to assign u
any of the grammatically possible metaphorical characters in its metaphor
set borders on linguistic incompetence.
B. I hear u, recognize that it has a possible metaphorical interpretation
but, given what else I know about the speaker (e.g., that he is hopelessly
literal minded and never uses metaphors), I fail to recognize that this is
how he intended it. Therefore, I fail to interpret u metaphorically. Here I
202 Chapter 6

rule out the possibility that u is to be, or might be, interpreted metaphor-
ically because of my other presuppositions. This failure is not linguistic,
but empirical.
C. I recognize that at least one expression in u is to be interpreted meta-
phorically but I do not know which one. Or perhaps I identify the wrong
constituent. For example, I fail to get your metaphor because, while we
both know that `S is P' (e.g., `the sun is smiling') is to be interpreted as a
metaphor, you interpret S literally and P metaphorically while I interpret
S metaphorically and P literally. This failure falls between the linguistic
and extralinguistic, depending on the reason that the correct constituent is
not identied.
2. Cases where we fail to know the context-set of presuppositions, and
hence the content of the metaphor in its context. This is failure to grasp
one kind of knowledge by metaphor.
Here we know that u is to be interpreted metaphorically. We know that
it is to be assigned a metaphorical character and we know how meta-
phorical character works. Hence there is no linguistic incompetence. But,
for whatever reason, we fail to learn the relevant presuppositions about its
m-associated features that determine the content of u in that context. This
failure is of extralinguistic knowledge, but we can dierentiate further
between cases.
Sometimes the interpreter is at a total loss to know the relevant pre-
suppositions. Ted Cohen's late father says `That beer is green' and I don't
have the foggiest idea of what associations with `green' and/or `beer' are
appropriate or relevant to interpreting his utterance. I can't identify the
presuppositions his father is making, and I can't come up with any of my
own. This is an extreme case of failure to know the context set of the
metaphor, and thus failure to understand its contentthough I know that
the utterance is a metaphor.
In a second set of cases, I identify a context set of presuppositions and
thereby assign u an interpretation. However, the interpretation diers
from the speaker's; hence I fail to understand how he intends u to be un-
derstood. And where I fail to realize that our context sets dier, I fail even
to realize that our interpretations dier.
This case is a frequent source of failure to get literary metaphors, where
the main interpretive task is recovering the relevant presuppositions. (I
won't take a position on what counts as legitimate evidence for a literary
interpretation; for our purposes both extratextual beliefs and intratextual
information can count as contextual information for the interpretation
of a given metaphor.) But it may not always be necessary for a correct
Metaphorical Character and Metaphorical Meaning 203

interpretation of a metaphor that I know or give the author's intended

interpretation (as some schools of literary criticism tell us). It may be
more important to give the most informative, interesting, illuminating
interpretation possible in the context, regardless of what the author in-
tended. To obviously dierent degrees, failure to do so may be counted by
some as a type of failure to ``get'' the metaphor.
3. Cases of failure to know the truth-value of the metaphorical utterance.
Here I correctly identify u as a metaphor, identify a context set of pre-
suppositions, know a content of the metaphor in the context, but then
either fail to know its truth-value (on the metaphorical interpretation) or
mistake the truth-value. Many philosophers would not count this failure
as a failure of understanding, but I think a case can be made that it is,
especially for those metaphors whose process of interpretation has a
strong constructive element. Suppose I interpret u with presuppositions
that yield a metaphorical interpretation that is obviously false or absurd.
Suppose the propositional interpretation is so obviously false or absurd
that if it were the literal meaning of an utterance, it would be possible
grounds to identify it as a metaphor. If I give the metaphor this absurd
interpretationwithout a collateral reason to so interpret it (because, say,
of some additional information communicated by its metaphorical mode
of expression)then one might argue that I have also failed to ``get'' the
metaphor. If I can't believe the irrational or absurd, I also can't interpret
someone as expressing the irrational or absurdor, if I so interpret
someone, at least the burden of responsibility falls on me to show that I
have not misinterpreted her.
4. Cases where we fail to know what metaphorical character is. In this
last class of cases, we may know that some utterance is not to be inter-
preted literally but not yet know what it is to interpret it metaphorically;
say, we do not know that a metaphor depends for its interpretation on a
contextual parameter (as opposed, say, to its literal meaning), or on a
particular contextual parameter (i.e., presuppositions) dierent from that
which determines whether its content is true (i.e., the actual circum-
stances), or only on its actual context set of presuppositions (and not the
features m-associated with the expression in a counterfactual context in
whose circumstances we also want to know whether what the metaphor
actually expresses is true). In these last sets of cases, we have not yet
mastered the function of metaphorical character. A speaker in this posi-
tion might be said to lack an ``entry'' for `Mthat' in his mental lexicon.
Although he knows that the utterance is not to be interpreted literally, he
is in no better position to interpret it metaphorically than he would be
204 Chapter 6

with respect to a string that he knows to be in a foreign language but of

which he does not know a single word.
Of these cases, the incompetence exemplied by the fourth is solely due
to linguistic incompetence; the lack of a kind of knowledge of language
proper, similar to the linguistic incompetence of one who does not know
that the content of (a token of ) the indexical `now' depends on its time of
utteranceand on the actual time rather than, say, the time the speaker
believes it to be. This fourth kind of incompetence is also best described as
lack of knowledge of meaning, not of content or truth-conditions and not
of the context, but of how one goes about interpreting a metaphor. This
case of linguistic incompetence in metaphor is also especially interesting
precisely because it is so hard to imagine. It asks us to conceive of a
speaker who is fully competent ``speaking literally,'' a full member of
our linguistic community, but incompetent ``speaking metaphorically,'' a
radical foreigner. From the point of view of our theory that draws such a
tight connection between metaphorical interpretation and other types of
context-dependent interpretation (e.g., of demonstratives), it also asks us
to think of a speaker who has the general repertoire for contextual inter-
pretation (assuming that is also part of his semantic competence in the
literal interpretation of his language) but who has somehow failed to re-
alize that metaphorical interpretation is of the same ilk. Without special
pleading, it is dicult to see how such a speaker could evolve.
Of the four cases, the source of most inter-interpreter disagreements
the kinds of hard-to-resolve disputes that make many philosophers skep-
tical about there being determinate metaphorical contentis the second,
although it may often appear that the disagreement is instead over truth-
value. Suppose, for example, that `Juliet is the sun' is asserted by Romeo
and denied by Paris. How can they settle their disagreement? Or, better
yet, what is the source of their disagreement? Romeo points out Juliet's
unparalleled qualities and her presence that inspires hope. Paris denies
neither that these properties are true of Juliet nor that they are true of the
sun. But Paris points out other properties of the sun: its remoteness or the
monotonous regularity and mathematical predictability of its motion.
That those properties are true of the sun and false of Juliet is, in turn,
admitted by Romeo. There is no disagreement between them, then, over
what is true either of the sun or of Juliet. Where they disagree is over
the question: Which of these individual context sets of presupposed m-
associated features ought to be adopted as the shared context set for the
metaphorical interpretation of `is the sun' on the occasion? Although they
Metaphorical Character and Metaphorical Meaning 205

appear to disagree over the truth-value of the utterance, in reality their

disagreement is over the proposition expressed. Insofar as they each hold
dierent interpretations of the metaphor in their respective individual
context sets, they are talking past one another, inhabiting dierent con-
texts, not disputing the truth-value of a single proposition expressed in a
shared context. To resolve their disagreement, Romeo and Paris must rst
agree on a shared context set of presuppositions. Only then can they agree
on the propositional content of the metaphor, and only having xed on a
common proposition can they raise the question of whether the facts
make it true.
A good, or even just adequate, theory of metaphor should have the
conceptual resources to articulate and explain the dierences between
these cases. But to do this, it is necessary to invoke the character-content
distinction, which presupposes a distinction between linguistic knowledge
and extralinguistic information, between meaning and the presuppositions
that ( jointly with the meaning) determine the content of the metaphor in
its context.

III The Interpretation vs. Evaluation of a Metaphor

The crux of our semantic theory lies in the distinction between the char-
acter and content of a metaphor or, equivalently, between the second and
third of the three (or, if you wish, four) stages of metaphorical compre-
hension and roles of the context distinguished in chapter 2: between the
interpretive role of the context, given the character of the metaphor, in
generating what is said, and its evaluative role, given a content generated
in a context, in determining the actual truth-value of the utterance. (For
this last role of evaluation, it is the circumstances of the context that do
the work, for the former role, the context set of presuppositions.) Without
this distinction, there would remain no part of the speaker's knowledge of
metaphorical interpretation that could count as part of his semantic
competence proper, or as knowledge of metaphorical meaning.
One strategy for arguing that we should draw this distinction for met-
aphor is by way of the same kind of argument that requires that we
distinguish character from content (and the corresponding roles of the
context) for demonstratives. In their case, you'll recall, Kaplan argues for
the distinction from the diculties that result either when we simply
combine contexts and circumstances in complex indices or when the rule
describing the character of a demonstrative is taken as its propositional
content.4 If pointing to Romeo, I say
206 Chapter 6

(1) That man might not have loved Juliet,

(1) will be true if and only if there is some counterfactual circumstance in
which Romeo, the actual referent of the demonstrative in its context of
utterance, does not love Juliet, not if there is some counterfactual cir-
cumstance in which whoever would be demonstrated there doesn't love
Juliet. Likewise, if I say
(2) I might not be speaking
my utterance is true if and only if there is some counterfactual circum-
stance in which I, JS, am not speaking, regardless of the fact that in that
circumstance I could (and would) not be xed there as the direct referent
of the indexical given its character. The character of the indexical func-
tions only to generate its content in a context; it plays no part in the
evaluation whether that content is true at any circumstance. Now, this
argument is especially forceful applied to singular demonstratives and
indexicals, for their context-dependent contents are the very individuals
who are their direct referents. But an analogous claim holds for the
properties that constitute the propositional contents of predicate demon-
stratives and of metaphors. The content of the predicate demonstrative
and of the metaphor is always the property xed in its context of utter-
ance. Suppose Paris disagrees with Romeo's utterance of (3)
(3) Juliet is the sun
in c but concedes that
(4) Juliet might have been the sun.
(4) is true just in case there is some counterfactual circumstance w 0 in
which Juliet has the particular set of properties P, which is the content of
`is the sun' interpreted metaphorically in its actual context of utterance c.
Juliet must fall in the extension at w 0 of the relevant property P, but the
relevant property P is xed relative to c, not w 0 . It is not sucient if there
is some counterfactual circumstance in which Juliet possesses whatever
property happens in that circumstance to be the content of `is the sun' had
it been interpreted metaphorically there. All that is relevant to determin-
ing whether (4) is true in its actual circumstance, w(c), is whether Juliet
has P at w 0 , regardless of the truth-value in w 0 of the proposition, if any,
that would have been expressed had the sentence been interpreted there at
w 0 . Likewise, it matters not at all whether the denizens of w 0 are able to
express the property using their linguistic resourceswhether the prop-
Metaphorical Character and Metaphorical Meaning 207

erty P, xed in c, is metaphorically expressed (or expressible) at w 0 by `is

the sun'.
If your intuitions agree with mine about this example, you should also
agree with the following claims. Since sentences that contain metaphors,
like demonstratives, express dierent propositional contents in dierent
contexts, we must distinguish between the content actually expressed by
such a sentence in its context of utterance and the content that it would
have expressed in some other context. We must also distinguish between
the truth-value of the proposition actually metaphorically expressed were
it evaluated at a counterfactual circumstance and the truth-value of the
proposition that would have been metaphorically expressed at that coun-
terfactual circumstance. And for both metaphors and demonstratives, the
relevant or appropriate propositional contentthat is, the proposition
asserted, to which the speaker's utterance commits himis always that
content expressed in its context of utterance, regardless of the circum-
stance at which we evaluate its truth.
One prima facie counterexample to these thesesand, in particular, to
the thesis that a metaphor is always interpreted relative to its actual con-
textis the occurrence of a metaphor in counterfactual statements. The
context relative to which we interpret the metaphorical expression
`Mthat[`the Rolls-Royce of ']' in the consequent of the subjunctive condi-
tional, (5):
(5) If Rolls-Royce starting building really rotten cars, SONY would be
the Rolls-Royce of television
cannot be its actual context of utterance, since we actually presuppose
hence the counterfactualthat a Rolls-Royce is the best car now made.5
But, in fact, these cases can be handled nicely by our account if we make
two additional (relatively innocuous) assumptions about the pragmatic
analysis of conditional statements.6
First, assume that the antecedent of a conditional is an explicit suppo-
sition that is added (at least for the duration necessary to evaluate the
consequent) to the initial context set of presuppositions. Second, assume
that the subjunctive mood in English is a conventional device for indi-
cating that prior presuppositions to the contrary are being (temporarily)
suspended. Here, then, we add the presupposition that Rolls-Royce has
started building rotten cars to our initial context (set) c and simulta-
neously ``contract'' c by dropping the initial presuppositions that associ-
ate features of excellence with the name `Rolls-Royce'. Call the resulting
208 Chapter 6

context (set) c*. This context, c*, not the original (actual) context c, is
now the actual contextthe context in which the consequent of (5) occurs
and is interpretedjust as our theory predicts. As we said in chapter 4,
section II, the members of the context set are presuppositions, not because
they are presupposed prior to the time of the utterance, but because they
are conditions required for the interpretation of the utterance.
A potentially more problematic philosophical assumption is implicit in
our claim that the property relevant to evaluating the truth of a meta-
phorical sentence at an arbitrary circumstance is always that property
actually expressed by the metaphor in its context of utterance. This claim
assumes that it is possible to identify the property at counterfactual cir-
cumstances other than the actual circumstance in which it is generated.
Here we might distinguish two cases. Where the property is rigid, always
yielding the same extension at all possible circumstances, this problem
reduces to that of cross-world identication of individuals (or sets of
individuals). But, in the second case, where the property is not rigid and
its extension itself changes over circumstances, how do we determine the
property that is reidentied? How are such properties individuated?
It is important to distinguish semantic from metaphysical issues raised
by this question. The property P that is actually expressed by one ex-
pression interpreted metaphorically may not be expressed by that very
same expression interpreted metaphorically in a counterfactual circum-
stance w 0 . We might also lack a context-independent (or nonmeta-
phorical) expression for P that would enable us to express that property
were we in w 0 . Nonetheless, even if we cannot actually express P except
by using a predicate sensitive to its context, the semantic question of
whether we can express the property at w 0 is independent of the meta-
physical question of whether the property P exists in w 0 . The latter is the
question whether P can be evaluated at w 0 . Evaluation is independent of
expression.
Evaluation of a property, in turn, raises its own set of questions. Is it
meaningful to talk of evaluating a property, or of the property ``existing,''
at a counterfactual circumstance even if the property applies to no indi-
vidual at that circumstance? Does the property itself, then, not apply (re-
fer, denote)hence ``exist''at that circumstance? Or is it simply the
case that the thing that it purportedly applies to does not exist in that
circumstance? (In the latter case, choose your favorite theory of non-
referring terms.) This issue is not specic to metaphor. Indeed it is the
same problem that arises for nondenoting singular terms and, among
them, nondenoting directly referential terms. So, choose your own favor-
Metaphorical Character and Metaphorical Meaning 209

ite theory to handle the whole set of cases but, whatever your line on this
question, keep the semantic and metaphysical issues separate. The former
is our exclusive concern here.
A last set of cases that illustrates the importance of the distinction be-
tween interpretation and evaluation are metaphors that occur in the
propositional attitudes, for example, belief reports and indirect dis-
course.7 Because I do not have a theory of belief to propose, a full
account of these cases is out of the question here. However, I would like
to point out certain parallels between the behaviors of metaphors and
demonstratives in these contexts; whatever ultimate account works for the
latter should also work (unless proven otherwise) for the former. It would
also be helpful, for a start, to distinguish two sets of questions raised by
metaphors in these linguistic constructions. The rst concerns the indi-
viduation of their truth-conditions; the second concerns whether their
truth-conditions exhaust the information they convey: their knowledge by
metaphor. I shall return to the second of these in chapter 7; here I'll
address their truth-conditions.
Under what conditions can I truly state (6)?
(6) Romeo asserted (believes) that Juliet is the sun.
There are at least four dierent interpretations or readings of this sen-
tence. On the rst and perhaps least problematic reading, the metaphor in
(6) is only the speaker's; the metaphor is simply his way of expressing a
property in his own context (i.e., relative to his own presupposition set).8
The speaker intends to attribute this property (as a constituent of a
believed or asserted proposition) to the subject (Romeo) but without
imputing that the subject would himself have expressed that property in
those metaphorical terms. According to that reading:
(6a) (b P) (P {Mthat[`is the sun']} (cs ) & R asserted (believes) the
proposition h j, Pi)
where cs is the speaker's context (set of presuppositions), the context of
utterance of (6). Apart from the fact that this involves quantication in
(of the second-order quantier), there is nothing problematic about this
reading.
On the next three readings, the metaphor used is ascribed to the subject.
That is, what is ascribed to Romeo is not only what he believes or
assertedthe (propositional) content of his beliefbut also how he
believes or asserted that content, its metaphorical character.9 Each of
these three readings must also solve one problem. Because metaphorical
210 Chapter 6

character is nonconstant, its content will vary across context sets of pre-
suppositions. Whose context is, then, the relevant context for the inter-
pretation of metaphors in indirect discourse and the attitudes?
Two dierent assumptions prima facie come into tension here. On the
one hand, we presume that the content expressed both in indirect dis-
course and in belief reports is the same as that of the subject's (reportee's,
e.g., Romeo's) original utterance and belief in their respective contexts.
But because that content is expressed by a (indeed the same) context-
sensitive expression (such as a metaphor) with one character, the context
of utterance of (6) must be that of the original utterance. On the other
hand, there is also the general constraint on the interpretation of index-
icals and demonstratives according to which their relevant context of in-
terpretationthe context to which their characters applyis always their
context of utterance, the context of the speaker of the indirect-discourse
sentence or belief report, no matter how deeply embedded the indexical or
demonstrative may be.10 The problem is to resolve, or accommodate,
these prima facie incompatible desiderata.
To give an example, suppose Romeo asserts (repeated here)
(3) Juliet is the sun
whose logical form is
(3a) Juliet Mthat[`is the sun'].
By disquotation and standard principles of indirect discourse and belief-
reporting, I say (6), whose logical form therefore is
(6b) Romeo asserted (believes) that Juliet Mthat[`is the sun'].
Both (3a) and (6b) contain a metaphorical expression with a nonconstant
character. Under what conditions is (6b) true?
There is one sense of `said' (and perhaps `believe') according to which
what is said is (only) the sentence (under an assignment of a character)
even where the sentence contains an indexicalrather than its content (in
a context). So, if two politicians say `I am the best presidential candidate',
in this sense of `say' they said the same thing. Similarly, both Romeo and
Paris might be said to say the same thing when they each utter `Juliet is
the sun' despite the fact that the presuppositions they respectively associ-
ate with `the sun' are dierent, even incompatible, yielding dierent, even
incompatible, contents in their respective contexts. In general, however,
this is not the relevant sense of `said' (or `believe') in indirect discourse or
belief reports. If Russell tells Frege
Metaphorical Character and Metaphorical Meaning 211

(7) You are wrong

and Frege reports this fact, he cannot say
(8) Russell said (that) you are wrong.
He must say:
(9) Russell said (that) I am wrong.
Here the reporting sentenceto be a reportmust preserve the content
of the original utterance in its context. And because the content (or direct
referent) of the indexical is always determined relative to its actual context
of utterance, the shift in contexts, or contextual parameters for `you' and
`I', forces a corresponding shift in the expressions used with their respec-
tive characters.
In the case of indexicals this is especially clear, for, as Frege (1984)
originally noted, there are systematic rules according to which we convert
and replace one indexical by another in order to preserve the content of
one utterance in a dierent context. In the case of metaphorsdespite
their parametricitythere are no analogous rules that enable us me-
chanically to replace a metaphor with one character by another in order
to preserve the content of the original across variations in the relevant
contextual parameter. But, again, metaphors are not unique among
context-dependent expressions in this respect. There are no xed, me-
chanical rules for complete demonstratives, complex demonstratives, and
dthat-descriptions that legislate how one of these should be replaced by
another to preserve the original content in a dierent context. Yet, the
same constraint applies: Their contents (e.g., direct referents) are always
determined relative to their contexts of utterance, no matter how deeply
embedded they are in indirect discourse or attitudinal operators. The
speaker's context, not that of the subject of report, is what matters.
The only way to explicitly and unequivocally ensure that the reported
utterance, preserving the original sentence uttered, is interpreted relative
to the subject's or reportee's original context of utterance, is to quote it
directly:
(6c) Romeo asserted (believes) (that): `Juliet is the sun',
where what is quoted is not just an uninterpreted expression but the ex-
pression with its original interpretation. And perhaps some uses of indi-
rect discourse border on direct discourse uses, thereby shifting the relevant
context of interpretation from the context of actual utterance to the orig-
inal context of the reported utterance. Bracketing these cases, however,
212 Chapter 6

let's return to our original question: Under what conditions will (6) be
true?
The problem, as we said earlier, is that if we try to preserve both the
character and content of the original reported utterance, we violate the
constraint that the interpretation of the metaphorical expression, like that
of indexicals and demonstratives, always cleaves to its context of utter-
ance, the context of the speaker, not of the subject of the report. (I'll call
that condition the actual context constraint.) So, if we analyze (6) as
expressing what it presumably does, namely,
(6d) R asserted (believed) the proposition h j, {Mthat[`is the sun']}(cr )i
where cr is Romeo's context (set of presuppositions), not the speaker's, we
violate the actual context constraint. On the other hand, if we interpret
(6) as (6e)
(6e) R asserted (believed) the proposition h j, {Mthat[`is the sun']}(cjs )i
attributing to Romeo the proposition generated by my (JS), the speaker's,
presuppositions cjs , we respect the constraint but attribute to Romeo a
proposition he may never have held, given his dierent presuppositions
associated with the expression interpreted metaphorically.
One option lies between merely (directly) quoting the original utter-
ance's character and generating a full-blown content that would require
shifting to a context other than the actual one. I may truly report
Romeo's original utterance if I merely claim that he said something, rela-
tive to his presuppositions, using the same words (with the same charac-
ter). Thus (6f ):
(6f ) (b P) (b c) (c Romeo's context set of presuppositions and
P {Mthat[`is the sun']}(c) and R asserted (believed) the proposition
h j, Pi)
If we want to do more, to ascribe a specic, fully determined content to
Romeo, either we can preserve the original content at the expense of the
character, or we can try to replicate the context c* of the original utter-
ance by making our (the speaker's) context of utterance c suciently
similar to c* in all respects relevant to determination of the content of the
original utterance given its original character. On the rst alternative, we
abstract the content away from the context; on the second, we put our-
selves in, or take the point of view of, the original context, yielding the
same value for the same character.11 Let me say a bit more about what
``putting ourselves in'' or ``taking the point of view of '' the original con-
text might involve.
Metaphorical Character and Metaphorical Meaning 213

Suppose I report
(10) Shakespeare said that the world is a stage
which, in its original context of utterance (reconstructed from As You
Like It) expressed, say, that the men and women of this world seem to
follow rehearsed patterns of action throughout the course of their lives,
patterns that divide into distinct parts, with beginnings and ends like the
acts of a play. And suppose that I have rather dierent presuppositions,
according to which the interpretation of the metaphor is that life and
action in this world is all pretense and illusion, that humans are all vain
creatures and their actions mere spectacle.
Not only do I intend to report Shakespeare's utterance, and not only do
I use his metaphor to express its content; my use of the linguistic con-
struction for reporting thereby commits me to using the metaphor with
his intentions and presuppositionseven where that commitment re-
quires that I suspend the presuppositions I myself would associate with
that metaphor. After all, to disregard the subject's presuppositions for my
own, the reporter's, would defeat the objective of giving a report. There-
fore, on pain of either acting irrationally (contrary to my own desire to
report his belief ) or exposing a kind of linguistic incompetence on my part
(with respect to either or both metaphor and belief reports) I must adopt
the reportee's point of view. There may be an appearance that the content
of the metaphor is determined relative to the context of the subject of the
report rather than relative to that of the speaker, the actual context of
utterance. But because the speaker must suspend the presuppositions
he otherwise, by default, would hold in order to adopt the local pre-
suppositions of his subject, in fact the metaphor is interpreted relative to
its speaker's presuppositions, albeit his adopted rather than native ones;
nonetheless relative to what is for the utterance its actual context. The
actual context constraint is upheld.
At the same time, for a speaker to enlist presuppositions by ``taking the
subject's point of view'' is evidently not the same as making those pre-
suppositions himself. Taking the subject's point of view is not simply a
matter of adding presuppositions to my own context set, even tempo-
rarily. Adopted presuppositions are insulated from the speaker's general
context set of presuppositions. What holds according to the adopted pre-
suppositions (say, in the circumstances that would obtain were the pre-
suppositions true) cannot be assumed to hold according to the speaker's
native presuppositions.12 Hence, although in general we do not assert
anything that contradicts the presuppositions of our prior assertions,
214 Chapter 6

there is nothing to prevent us from explicitly denying the adopted pre-

suppositions of a belief report or indirect discourse, for example,
(11) Shakespeare said that the world is a stage, but our lives are nothing
but sheer chaos.
Perhaps with metaphors in belief reports, we could also say, then, that
pretense or pretending is at work in the relevant attitude of presupposi-
tiondespite our general reservations expressed in chapter 4. But note
that this is not specic to metaphors in belief reports; it is true of all belief
reports. We can also cancel nonmetaphorical presuppositions adopted to
take the subject's point of view:
(12) Bill said that Mary's husband stopped beating her, but Mary is not
married and she is a well-known masochist.
This brief discussion suggests that although adopted presuppositions
are not native, they are no less actual, that is, held (if only in a slightly
broader sense) in the speaker's context of utterance. Where the speaker
enlists such presuppositions, it is not necessary to hold that the interpre-
tation of the utterance shifts to a counterfactual context, a dierent con-
text than that of the speaker. Therefore we can still maintain that the
general features that characterize demonstratives and indexicals apply
also to the interpretation and evaluation of a metaphor. Indeed the sim-
plest explanation why metaphors cleave so closely to their contexts is that,
just like demonstratives and indexicals, they are distinguished by their
nonconstant character.

IV Metaphorical Meaning

The various arguments for the character-content, or interpretation-

evaluation, distinction reviewed in section III constitute one way to argue
for the claim that there is metaphorical meaning, that there is knowledge
of meaning specic to metaphors, apart from their contents (or truth-
conditions) in particular contexts. But there is also a second way to argue
for metaphorical meaning. Harking back to our discussion of Davidson,
one might argue that precisely because metaphors are highly context-
sensitive and context-variable, they need meanings to constrain their
possible interpretations. That is, meanings are constraints on possible
contents that display their structure. The actual contents (truth-conditions)
of a metaphor are never part of its meaning, and the conditions that con-
stitute the meaning of the metaphor are never part of its content (truth-
Metaphorical Character and Metaphorical Meaning 215

conditions) in any context. But the twomeaning and contentare

intimately related. The task of the metaphorical meaning of (a token of )
an expression is rather to constrain the possible content the expression
expresses qua metaphor in a context.
For example, I argued in chapter 2 that there are constraints on the
possible metaphorical interpretations that can be assigned to verb phrase
anaphors jointly with their antecedents. These constraints also seem to be
identical to those that apply to nonmetaphorical verb phrase anaphora,
where the antecedent allows multiple interpretations but only the inter-
pretation it is actually given can serve as the interpretation of the ana-
phor. Thus, recall that `may' can be interpreted with the sense of either
permission or possibility in
(13) John may leave tomorrow
but in
(14) John may leave tomorrow, and Harry, too
the antecedent and anaphor must both be interpreted with the same con-
tent: Either both must mean permission or both possibility. The explana-
tion for that constraint is that in cases of anaphora the interpretation of
the antecedent is copied onto the anaphor; hence, although the interpre-
tation of the antecedent may be free, once interpreted, its anaphor will be
obligatorily assigned the same interpretation.
Similarly with metaphorical antecedents and anaphors. Consider, rst,
the literal statement
(15) The largest blob of gases in the solar system is the sun
and the two (I shall assume) dierent metaphorical interpretations of `is
the sun' in (repeated)
(16) Juliet is the sun
and
(17) Achilles is the sun.
(The substantive dierences between the two interpretations need not
concern us, so long as there is a dierence.) Now, consider these (unac-
ceptable) examples of verb phrase anaphora:
(18) *The largest blob of gases in the solar system is the sun, and Juliet
is, too
(19) */?Juliet is the sun, and Achilles is, too.
216 Chapter 6

On our account, the explicit representations underlying (16) and (17) and
their respective propositions (in their contexts) are
(16*) Juliet Mthat [`is the sun'].
h j, {Mthat [`is the sun']}(c1 )i
and
(17*) Achilles Mthat [`is the sun'].
ha, {Mthat [`is the sun']}(c2 )i
Although the two occurrences of `is the sun' have the same character,
they have dierent contents in their respective contexts. Hence the prop-
ositions expressed by the antecedents and anaphors of (18) and (19) are
(18*) hhThe denite description operatorhBeing-a-blob-larger-than-any-
other . . .ii, Being-the-suni&h j, {Mthat [`is the sun']}(c1 )ii
(19*) hh j, {Mthat [`is the sun']}(c1 )i&ha, {Mthat [`is the sun']}(c2 )ii
We can now explain why (18) and (19) are unacceptable, although (18) is
still worse than (19). In (18) the purported antecedent and anaphor are
expressions with dierent characters; failing to stand in an anaphoric re-
lation, the ``meaning,'' or character, of the rst cannot be copied onto the
second. In (19) the antecedent and anaphor do have the same character
but have dierent contents; hence, as marked, they must occur in dierent
contexts. This raises two problems. First, in utterances like (19), the
statement must ``shift'' contexts in the course of its interpretation, split-
ting itself between two contexts. This is at odds with a general assumption
that complete sentences are always interpreted relative to a single context.
For the same reason, (20) will be unacceptable where each conjunct has
the same interpretation it would have in isolation.
(20) *Juliet is the sun and Achilles is the sun.
Here too we must assume that the interpretation of the string as a whole
requires context-shifting midway. Second, the character of the antecedent,
but not its content, can be copied onto the anaphor, suggesting that only
where there is full identity of character and content can the copying that
underlies anaphora completely work. Where there is less than complete
identity between character and content, the interpretation of the anaphor
is blocked, rendering the result unacceptable (at least to a degree).
To explain these constraints on interpretation, it is necessary, then, to
appeal to the notions of character and content. Since the dierences in
interpretation clearly make a dierence to the truth-values of the sen-
Metaphorical Character and Metaphorical Meaning 217

tences, the constraints are semantic. And if meaning is what captures

semantic constraints, this is to say that there is meaning specic to the
metaphorical interpretations of metaphors.
A similar explanation applies to the failure of substitutivity discussed in
chapter 2. Suppose that Romeo utters (to repeat)
(21) Juliet is the sun
in a context c in which `is the sun' is interpreted metaphorically to express
the property P of being unequaled by one's peers. Suppose also that (21),
so interpreted, is true in the circumstance of its context, namely, the
actual world, w(c). But
(22) The largest gaseous blob in the solar system is identical with the
sun
interpreted literally, is also true in w(c). Yet, from (21) and (22) it does
not follow that
(23) Juliet is the largest gaseous blob in the solar system
will be true in w(c) even if `is the largest gaseous blob in the solar system'
is interpreted metaphorically (to say, e.g., that she is liable to explode at
any moment).
Here the explicit representation underlying (21) is more precisely
(21*) Juliet Mthat [`is the sun'].
The representation underlying (22) is the same as its surface form, but
(23), like (21), ought to be represented as
(23*) Juliet Mthat[`is the largest gaseous blob in the solar system'].
Thus (21*) and (23*) involve dierent (though homonymous) expressions
than (22)that is, expressions with dierent characters, and what is nec-
essary for substitutivity is that the substituted and substituting expressions
be both character- and content-equivalent. It should also now be clear
why, as I rst conjectured in chapter 2, one might be tempted to propose
that the explanation for the failure of substitutivity is something like a
fallacy of equivocation. However, as I also argued in chapter 2, the met-
aphorical interpretation of an expression is not just an additional sense of
a polysemous expression. Nor does metaphor t neatly into the received
typologies of ambiguity. The proposed analysis of metaphorical interpre-
tation in terms of character, or meaning, and content explains why.
Finally, the character-content distinction provides an elegant solution
to a classical problem about metaphorical meaning. Most of us (unless we
218 Chapter 6

are philosophers with jaded conceptions of truth) share the intuition (or
act as if it is the case) that utterances of declarative sentences containing
metaphors are truth-valued. When two people argue over a claim ex-
pressed by a sentence containing a metaphor, one of them is correct, the
other incorrect; there is a ``fact of the matter'' (at least as much as with
literally interpreted sentences) to which metaphorical statements are ac-
countable. So, for example, when Paris disagrees with Romeo's utterance
`Juliet is the sun', they are disputing the truth of Romeo's statement.
(Which is not to say that they could not also be disputing the value of
expressing that claim by way of that metaphor, as I'll argue in ch. 7.
However, that dispute is additional to their disagreement over the truth of
the claim.) The classical problem this intuition raises is this: What is it in
these utterances of metaphor that is true or false? If metaphors are truth-
valued, what is the vehicle that bears their truth-value?
The most obvious candidate is the sentence. But if we individuate a
sentence by its ``meaning'' (i.e., those aspects of its interpretation that are
a function of its linguistic form), the relevant truth-bearer cannot be the
sentence unless we also specify its meaning or, more precisely, which
meaning it has on that occasion. It clearly cannot be the sentence with its
literal meaning because many metaphors are, as a matter of fact, literally
false but metaphorically true. Hence taking the literal sentence as vehicle
yields the wrong truth-value.
The truth-vehicle also cannot be the sentence with its metaphorical
meaning. First, as in the literal case, the same sentence with one meta-
phorical meaning can express dierent contents with dierent truth-values
in dierent contexts. Hence it would be necessary rst to specify the rele-
vant context. Second, this candidate raises a slew of diculties having to
do with shift and change of meaning. If the claim assumes that the meta-
phorical interpretation (or use) of the sentence is the meaning of the
sentence, that its metaphorical meaning is the only meaning of the sen-
tencehence, that it ``loses'' its literal meaning as it ``gains'' the meta-
phorical onethen the claim is at best highly implausible. If that is the
case, every time a speaker uses a word metaphorically to express some-
thing other than its literal meaning, the word would change its meaning.
(Deconstructionists may indeed hold a view like this, but such a vacuous
notion of meaning has no explanatory power.) Furthermore, if the met-
aphor loses its literal meaning, we have lost the means to explainex-
cept historically or diachronicallyhow the metaphorical interpretation
depends on the literal meaning of the word, a relation central to our idea
of metaphorical meaning/interpretation. If the metaphorical interpreta-
Metaphorical Character and Metaphorical Meaning 219

tion of an expression F depends (synchronically) on its literal meaning, F

must ``have'' (in some sense) that literal meaning even while it is inter-
preted metaphorically. But if the sentence ``has'' its literal meaning, we
are back at our original problem of identifying the relevant truth-vehicle.
We need something, then, to serve as the truth-vehicle for metaphors,
and none of the usual candidates will do. One philosopher who appre-
ciates the diculty is Robert Fogelin, who wonders ``how it could be
possible for an utterance to be false when taken literally, but true when
taken metaphorically without there being any shift of meaning from the
one reading to the other'' (1988, 91). In reply, he
rst note[s] that parallel situations arise with non-gurative language. If I say ``I
am James Jones,'' and James Jones says ``I am James Jones,'' then, unless I hap-
pen to be named James Jones, what I say is false and what he says is true, even
though there is no shift in the meanings of the words used. Less controversially,
and closer to the present case, if `x is good' means something like `x satises rele-
vant standards of evaluation,' then saying that x is good could be true in some
contexts, but not in others, without there being any shift in the meaning of what is
said. This, it seems to me, is how likeness claimsboth literal and gurative
function: what they say is that one thing is like another, and whether that's true or
not will depend upon canons of similarity determined by the context.13

There is much in this passage on the mark: the analogy with indexicals,
the implicit relativization of the attributive adjective `good' to a reference
class, the context-dependence of similarity judgments. But it is dicult to
see how Fogelin puts these ingredients together to solve the problem with
which he started: the problem of identifying the truth-bearersentence,
meaning, or whateverof sentences containing metaphors. Fogelin's
claim is that ``a metaphorical utterance of the form `A is a F' just means,
and literally means, that A is like a F''; yet both the metaphor and its
corresponding elliptical simile can be true in one context and not another,
not because their meaning changes, but because the ``modes of relevance
and evaluation governing the likeness claim'' (ibid., 7576) shift.
Now, as I argued in chapter 5, similarity judgments (that depend on
criteria like salience) are context-dependent. But it does not solve the
original problem of the identity of the truth-bearer to say that the ``canons
of similarity'' shift from context to context. The truth-bearers are obvi-
ously not the canons themselves. Nor is the problem solved by the sug-
gested parallel to sentences of the form `x is good'. Even if the semantics
of the latter were analogous to that of metaphor, there would remain the
same question about the identity of the truth-vehicle for utterances of
sentences of that form.14 Finally, because the modes of evaluation of the
220 Chapter 6

truth of likeness claims shift across circumstances (including the circum-

stances of contexts of utterance), it also cannot be the corresponding
likeness sentence `A is like a F' with its invariant meaning but shifting
``modes of relevance and evaluation'' that is the truth-bearer for the met-
aphorical statement `A is a F'. Such an explanation, once again, runs
afoul of the distinction between contexts of interpretation and circum-
stances of evaluation. Contrast the following modal metaphorical and
likeness sentences whose respective evaluations make reference to alter-
native circumstances, circumstances distinct from those of the contexts in
which their respective interpretations are generated. If (as Fogelin claims)
`A is a F' (where F is interpreted metaphorically) means that A is like a
F, it follows that `A might be a F' (where F is interpreted metaphorically)
should just mean that A might be like a F. But `A might be a F' uttered
in c is true in c just in case there is some alternative world or context c 0 in
which A has whatever property is expressed by F in c (say, the way in
which A is like F in c). In contrast, the sentence `A might be like a F' is
true uttered in c just in case there is some world or context c 0 in which A
has whatever property in virtue of which it is like F in c 0 . Here the meta-
phor and its respective likeness-statement clearly have dierent truth-
conditions, a dierence that emerges only when we separate (as in these
modal linguistic contexts) the tasks of interpretation and evaluation.
Fogelin's own conclusion notwithstanding, it is in fact his rst parallel,
to indexical sentences, that is ``less controversial . . . and much closer'' to
metaphor. Furthermore, I would conjecture that Fogelin correctly sees
what needs to be argued in order to solve the dilemma he raises, but
lacking adequate semantic distinctionsin particular, something like the
character-content distinctionhe cannot articulate the solution.15 Just as
the meaning of the indexical sentence `I am James Jones' does not change
when it is spoken truly by Jones and spoken falsely by me, so with meta-
phors. But what that shows is that there must be a distinct truth-bearer,
``what is said,'' that is neither a meaning nor a sentence. In other words,
we must distinguish the propositional content of the indexical or meta-
phorwhich is the truth-bearerfrom its character; we must distinguish
what is evaluated from what determines the interpretation.
Only if we distinguish the dierent elementscontext (or contextual
parameter, e.g., similarity- or salience-presuppositions), character, and
contentcan we adequately solve the problem of the identity of the truth-
bearer for metaphors. The explanation on my account of how it is possi-
ble for an utterance to be false when taken literally and true when taken
metaphorically without a shift of meaning is straightforward: The dierent
Metaphorical Character and Metaphorical Meaning 221

truth-values are a function of the dierent contents of the utterances (in

their respective contexts), and contents are not meanings. Characters
are meanings. As with indexicals, one meaning/character of a context-
sensitive expression can determine dierent contents in dierent contexts,
and the same content can be determined even in the same context by dif-
ferent meanings/characters. Hence a dierence in content does not entail
a dierence in the meanings/characters of the utterances.16 On the other
hand, on my account Romeo's utterance of `Juliet is the sun' does have a
dierent meaning on its metaphorical interpretation (under which it is
true) than it has on its literal interpretation (under which it is false). But it
would be misleading to describe this dierence as a shift in the meaning of
a single expression (type). The type `is the sun' (the type the utterance
would be assigned were it interpreted literally) itself undergoes no change
of meaning; it retains the (literal) meaning/character it always had. What
is dierent is that, when the utterance is interpreted metaphorically, it, the
utterance, is assigned to a dierent type, namely, the type of the meta-
phorical expression `Mthat[`is the sun']' whose character/meaning is dif-
ferent from that of `is the sun'. What changes is not the character of the
expression `is the sun' but the expression to be identied with the words `is
the sun'although, with dierence in expression (type), there is also a
corresponding dierence in character.
I can also now explain how, according to my account, the metaphorical
depends on the literal. I have already argued that neither the content nor
extension (in its context) of the metaphorical expression `Mthat[F]' is de-
termined compositionally from the content or extension (in its context),
respectively, of its constituent F. Metaphorical expressions, like index-
icals, are parametric. Yet, notwithstanding the fact that they are seman-
tically simple, they may be lexically complex. The characters of `Mthat[`is
the sun']' and `is the sun' are dierent, but the former is ``built up'' from
the latter. Metaphorical expressions are individuated by the characters of
their constituent expressions, and the relevant presuppositions are those
m-associated with their literal vehicles, the component expressions under
their literal characters/meanings. Therefore, there is a change of char-
acter when a word is interpreted metaphorically rather than literally, but
the vehicle interpreted metaphorically synchronically retains the literal
meaning on which its metaphorical meaning, or interpretation, depends.
This way in which the metaphor still has its literal meaning should not be
taken to mean (contra Davidson, as I argued in ch. 2) that the vehicle
is used literally, that is, interpreted and understood with that literal
meaning. Although the literal meaning of the vehicle plays a role in
222 Chapter 6

determining, and individuating, the character of the expression under its

metaphorical interpretation, the vehicle is not itself understood, or inter-
preted, literally on the occasion of its metaphorical interpretation or use.
If it were, then we should also say that the literal sentence, say, `Juliet is
the sun' is understood, or interpreted, literally on the occasion. But, as I
argued (in ch. 2, sec. III) following Margalit-Goldlum and White, it is far
from clear that we have an understanding of this literally interpreted sen-
tence, that we know under what conditions it would be true. The dier-
ence in questionbetween the vehicle having its literal meaning and
being understood, or interpreted, literallyis marked in my formal con-
struction of metaphorical expressions in which, it should be noted, the
vehicle, or word interpreted literally (the operand), occurs in single
quotes, metalinguistically. This is not Tarski/Quine-quotation (in which
there is no internal structure to the complete quotation and what is men-
tioned is an uninterpreted word); think of this kind of quotation as dis-
playing its value, which is identical to the word-token (with its literal
meaning) occurring in the metaphorical expression.17 Even while it is not
being used literally, that value has its literal meaning, which is necessary
to determine, or individuate, the character of the complete metaphorical
expression. It is this special intermediate standing of the literalaccord-
ing to which the metaphorical depends on the literalthat we sought to
capture in chapter 2. The same standing is, I believe, what some authors
are attempting to pin down when they describe a ``metalinguistic'' aspect
of metaphor in which we attend to the literal meaning of the word.18 But
it is only with the character-content distinction that we can clearly artic-
ulate the idea.

V Metaphor and Indirect Speech Acts

Yet another way to argue for the character-content distinction and the
claim that a metaphor's meaning is its character is through an analysis of
indirect speech acts performed by utterances containing metaphors. Sup-
pose, for example, that one Trojan warns another to ee the battleeld by
saying `the sun is blazing today', with the content that Achilles is furiously
angry. Or suppose I say: `this room is an icebox' on a wintry morning in
order to ask you to close an open window. How should we account for
the meaning(s) of these utterances?
On the standard explanation, the meaning of an utterance as an indi-
rect speech act is a kind of speaker's meaningwhat the speaker means
by uttering the words or sentencewhich is communicated in addition to
Metaphorical Character and Metaphorical Meaning 223

(although clearly related to) its sentence meaningwhat the words or

sentence means. Furthermore, in most cases the sentence meaning is its
literal meaning. For example, when I say `can you please pass the salt', I
am not asking a question about your ability but making a request; none-
theless the sentence uttered retains its literal meaning, which is supple-
mented by the meaning of the request. In John Searle's (1993) words: in
indirect speech acts, ``the speaker means what he says but he means
something more as well. Thus an utterance meaning includes sentence
meaning, but extends beyond it'' (110). With metaphors, however, the
story is more complicated. According to most use or speech act theorists
like Searle (as we saw in ch. 2, sec. I), metaphorical meaning is also a
variety of speaker's meaning.19 The speaker says S is P (whose sentence
meaning is that S is P), but he metaphorically means (as its utterance
meaning) that S is R, where R is related to P by one of the variety of
``principles'' we reviewed back in chapter 2. What is the relation between
these two kinds of utterance or speaker's meaning when a metaphor is
used to perform an indirect speech act?
Commenting on Searle's analysis, Robert Fogelin notes that it is one
question whether an utterance is to be interpreted literally or not; it is
another question whether the speaker's utterance meaning (or the ``point
of the utterance'') is ``exhausted'' by the sentence meaning or whether the
utterance meaning contains more. Fogelin calls any speech act performed
with a sentence whose meaning is only, or exclusively, (determined by) the
sentence meaning direct. A speech act whose (utterance) meaning is not
exclusively determined by its sentence meaning is indirect. Since the lit-
eral/nonliteral and direct/indirect distinctions cross-classify, we can now
distinguish four possible types of speech acts. Fogelin (1988, 39) suggests
the following examples:
i. Literal direct: ``Saying `the cat is on the mat' just [i.e., exclusively]
meaning that the cat is on the mat.''
ii. Nonliteral direct: ``Reciting nonsense poetry.''
iii. Literal indirect act: ``Saying `this hike is longer than I remember''
meaning (primarily but not exclusively) that I need a rest.''
iv. Nonliteral indirect: ``Saying `You're a real friend,' meaning you're a
louse.''
Examples (i) and (iii) are unproblematic. However, (ii) and (iv), the two
cases of nonliteral interpretation, bring our diculty to the fore. Fogelin
(ibid.) argues that irony falls in (iv) because ``the speaker expects the re-
spondent to reject the actual utterance and replace it with another that
224 Chapter 6

corrects it or modies it in certain ways.'' He claims that Searle would

also locate metaphor in this category. Metaphorical meaning is utterance
rather than sentence meaning and ``we typically mean more than what we
actually say and it is this `more' that really matters.'' Fogelin himself,
however, claims that metaphor falls under (iii) because, on his com-
parativist theory of metaphor, the sentence is ``intended literally'' but ``the
point is largely indirect.''
Suppose that I utter the sentence `this hike is longer than I remember'
to a colleague after making my way, for the nth time in as many years of
job searches, through a tall pile of job applications. Here `hike' is inter-
preted metaphorically, but the primary (if not exclusive) ``point'' of the
utterance, as with its literal interpretation in (iii), is that I need a rest. It is
dicult to know how Fogelin would (or could) classify this utterance. If
he takes it to be literalas a comparison between hikes and the process of
reading job applicationshe has lost the means to capture a signicant
dierence between the cited example in (iii) and our example, namely, the
dierence between the metaphorical and the literal. Likewise, although
Searle would place our example in (iv), his reason would be that all
utterances of metaphors fall in that category. He cannot account for the
dierence between this use of the metaphor to perform a genuine indirect
speech act and uses of metaphors that are directthat exclusively assert
their metaphorical content.
An analogous problem arises, on Searle's and Fogelin's accounts, for
the treatment of nonsense poetry. Searle does not explicitly comment on
this case, and it is not clear what he would say. However, Fogelin explains
that he classies it under (ii), the nonliteral direct, despite the fact that his
rst (wrong) instinct was that nothing could fall in this category, for if someone
intentionally utters something without meaning it literally, then it would seem that
there must be something else he is literally trying to get across, else why produce
an utterance at all? Nonsense poetry is a counterexample to this claim. (1988, 40)
Notice, to begin with, that Fogelin shifts from talking of ``meaning [an
utterance] literally'' to ``literally trying to get [something else] across.'' It is
not clear what he means by the latter phrase, but I assume that the sorts
of things we ``try to get across'' with utterances, apart from expressing
(asserting?) their meanings or contents (their locutions), are their illocu-
tionary and perlocutionary eects. Those are neither literal nor nonliteral.
What is more signicant is Fogelin's ``rst (wrong) instinct'' to which
he refers in the above passage. Presumably the explanation for his instinct
is that he (and, for that matter, Searle) assumes that all sentence meaning
is literal meaning and that the literal meaning of a sentence is its truth-
Metaphorical Character and Metaphorical Meaning 225

condition (relativized to a context). Therefore, where a sentence is trans-

parently false (e.g., a category mistake), it is dicult to imagine that a
rational competent speaker could intentionally utter it directly, that is,
exclusively and exhaustively, to mean by the words used what, and only
what, the words themselves mean. But this is dicult to imagine only
because he assumes that the only sentence meaning is literal meaning.
Once we acknowledge the character-content distinction, and allow for
metaphorical character as the meaning of a metaphorical expression,
there is no diculty nding nonliteral (e.g., metaphorical) direct utter-
ances. Romeo's utterance of `Juliet is the sun' is one of them. Its sentence
meaning is the character, or (metaphorical) meaning, of `Juliet Mthat[`is
the sun']'. And Romeo wants simply to assert that meaning or, more
precisely, its content in its context without doing anything further with
his utterance. Similarly, to return to the fourth class of cases, with the
character-content distinction we can give, and explain, genuine cases of
indirect nonliteral acts. Leaving aside irony, our earlier examples `the sun
is blazing today' and `this room is an icebox' are nonliterally interpreted
utterances whose point is not to assert their content or truth-conditions
(as determined by their metaphorical character/meaning) but to ``get
across something else,'' a warning and a request, respectively.
In sum, with a richer semantics for metaphorsone that captures the
character-content distinction and identies their characters with their
meaningwe can not only represent their proper truth-vehicle but also
account for the speech acts we perform with them.20

VI Nominative Metaphors

Metaphors like
(24) The sun is furious
whose subject term `the sun', in nominative position, metaphorically
refers to Achilles prima facie pose an additional problem for my notion of
metaphorical character. In the case of a predicative metaphor F, its con-
tent (in a context) is the value of the character of its underlying meta-
phorical expression `Mthat[F]' in the context, namely, a set of properties.
But if all metaphorical characters are functions from context sets of pre-
suppositions to sets of properties, the character of the nominative meta-
phor `the sun' in (24)whose value in its context is the individual
Achilles, not a set of properties, indeed not even a set of properties
uniquely satised by Achilleswill not be metaphorical. On the other
226 Chapter 6

hand, if the character of `the sun' in (24) is a function from the context to
an individual, it is not clear how to represent the fact that it is also a mode
of metaphorical referencerather than, say, simply a Donnellan-like
referential denite description in which the speaker succeeds in referring
to someone even though the referent does not satisfy its denoting con-
ditions.21 Somehow we need to capture both facts: that the nominative
term is metaphorical and that its content is an individual. How can we
accomplish this?
To sketch a solution to this problem, let me rst review some basics
about nominative terms. When I utter
(25) The prime minister of Israel in 1998 was a former furniture
salesman
I presuppose that Israel had a unique prime minister in 1998, and I assert
that he was a former furniture salesman. But (25) is ambiguous or, more
precisely, can be used in either of two ways, and the same ambiguity
carries over to the presupposition. The subject term `the prime minister of
Israel in 1998' can contribute as its content to both the presupposition and
assertion either a denotational conceptual complex, namely, the complex
consisting of the one and only one individual with the unique property of
being prime minister of Israel in 1998, or the actual individual who in its
context was the prime minister of Israel in 1998, call him, Bibi (b). Thus
the content of (25) (ignoring tense) can be either the general (or purely
conceptualized) proposition
(25a) hThe P, Fi
(where P and F are properties) or the singular proposition
(25b) hb, Fi.
Likewise, an utterance of (25) will either
(i) presuppose that there exists a unique P and assert that he is F
or
(ii) presuppose that b exists and assert that b is F.
If we use `Dthat' to generate directly referential terms from otherwise
general denite descriptions, we might also represent the ambiguity in
terms of the dierent surface subject terms `The prime minister of Israel'
and `Dthat[`The prime minister of Israel']'.
Apart from this semantic distinction, there is a second pragmatic dis-
tinction with which it intersects (and with which it is sometimes confused):
Metaphorical Character and Metaphorical Meaning 227

the distinction between terms (either denotational or directly referential)

used teleologically (or referentially, in Donnellan's sense), with a particu-
lar individual in mind to whom we use the term to refer, and those used
nonteleologically (or attributively), to refer to whoever ts the descriptive
conditions of the term or whoever turns out to be referred to (e.g., who-
ever is the historical or causal source of the term). It is worth emphasizing
that the denotationaldirectly referential distinction cross-classies with
the teleological-nonteleological one: Both general descriptions and directly
referential terms like dthat-descriptions can be used teleologically, with
someone in mind, or nonteleologically, to refer to whoever it is to whom
the term refers.22
With this background, let's return to nominative metaphors. To begin
with, if we distinguish between teleological and nonteleological meta-
phors, then (contrary to what I said in setting up the problem) perhaps we
should allow for propositions like those generated by (24), which contain
as their constituent for the nominative metaphor a conceptual complex
whose properties, or descriptive conditions, are those expressed meta-
phorically. For example, one might use `the sun' to refer to whoever turns
out to have (uniquely) those properties it metaphorically expresses. On
this use, (24) would mean something like
(24a) The sun, whoever it/he/she may be, is furious.
We might represent the proposition it expresses as:
(24b) hThe S*, Fi,
whose constituents are all properties and logical expressions, and S*
stands for the properties metaphorically expressed by `the sun' (not
including the literally expressed property of being the sun). That use of
(24) can, in turn, be contrasted with its metaphorical but teleological use
(24c) ha, Fi,
where a is the individual Achilles (even if Achilles is the individual who
uniquely satises S* in the context). (24b) raises no problems beyond
those raised by predicative metaphors. Our problem cases are exemplied
by (24c). It is not clear how we can semantically represent the metaphor-
ical character of `the sun' while also allowing Achilles to be its proposi-
tional content.
Here are two possible solutions. The rst rests on the assertion-
presupposition distinction. When I utter (24), I presuppose that there
exists a unique sun and that he is furious. Or, more accurately, I presup-
228 Chapter 6

pose that there exists a unique thing that has the properties that are the
content of `Mthat[`is the sun']' in c, and I assert that it/he/she is furious.
Here the content of my assertion is the singular proposition (24c), which
does not reect its metaphorical mode of expression. However, the asser-
tion presupposes a proposition whose representation does convey its
metaphorical mode of expression:
(24d) hThere is exactly one thing that possesses {Mthat[`is the sun']}(c)i
A second solution employs a dthat-description. On this proposal, the
logical form of the sentence that generates (24c) is
(24e) Dthat[`The x: x Mthat[`is the sun']'] is furious.
Here we rst determine the content (and, assuming the sentence contains
no other indexical elements, thereby recover the metaphorical character)
of the constituent-denoting description whose uniquely denoting descrip-
tive conditions are the properties metaphorically expressed by `is the sun'.
We then let the individual uniquely denoted by those conditions in the
context of utterance be the propositional constituent for the containing
dthat-description.
Both of these proposals apply not only to denite descriptions but also
to proper names interpreted metaphorically in nominative position. We
can represent the metaphorical character of
(26) Khomeini is coming
where we refer with the proper name to the Dean, either as a predicative
metaphor in the sentence expressing its presupposition
(26a) There exists an x such that x {Mthat[`is a Khomeini']}(c)
or as part of a dthat-description
(26b) Dthat[`The x: x Mthat[`is a Khomeini']'].
Finally, on both proposals, the metaphorical content (a set of proper-
ties) of the nominative metaphor `the sun' is not itself an immediate con-
stituent of the content asserted by (24). But it has a place in the full
explanation of its semantics. On the rst alternative, we make the pre-
supposition that contains the metaphorical content as an immediate con-
stituent in the course of our assertion. And in making the presupposition,
we communicate the specically metaphorical character-istic information.
On the second alternative, the character of the assertion is built up from
the metaphorical character, again giving it a place in the communica-
tion. On both proposals, it might be added, predicative metaphors play a
Metaphorical Character and Metaphorical Meaning 229

role explanatorily prior to that of nominative metaphors. We explain the

semantics of metaphors in nominative position in terms of the ``same
metaphors'' in predicative positions. If either of the two proposed solu-
tions is on the right track, predicative metaphors are, then, the more basic
construction in terms of which the other is to be explained.23

VII Metaphor and Simile

Given the proposed semantic theory of metaphor, the next question is

whether the same account applies to all kinds of nonliteral interpretation,
including simile, irony, metonymy, synecdoche, meiosis, and understate-
ment. Can and should we treat all these varieties of guration as instances
of one kind of context-dependent semantic interpretation, which depends
on sets of presuppositions that dier from each other only in their re-
spective contents? Should we invent cousins of `Mthat' for each of these
tropes?
I will divide this question into two. In this section I'll address the spe-
cial relation between metaphor and simile; in the next section, I'll turn
to the more general question of whether metaphor is representative of a
single natural kind of gures.
The history of the relationship between metaphor and simile goes back
at least as far as Aristotle: ``The simile also is a metaphor; the dierence is
but slight'' (Rhetoric 1406b 20). Note that Aristotle does not say that
metaphors are similes, or statements of comparison, but that similes are
metaphors.24 As I said in chapter 5, this way of putting their relation is
just rightand I shall return to it. But most recent authors who have
appealed to the authority of Aristotle (or criticized him) have taken him
to hold the inverted thesis, that metaphors are (elliptic) similes or com-
parison statements. In turn, they interpret Aristotle's claim to mean either
that the metaphorical meaning of statements of the form `a is (an) F' is the
literal meaning of corresponding statements of the form `a is like (an) F',
or (on a more sophisticated reading) that the literal meaning of the
metaphor is the literal meaning of the simile or comparison statement and
the gurative meaning of the metaphor is the same as the gurative
meaning of the corresponding simile or comparison statement.25
In evaluating the claim that metaphors are (elliptic) similes, let's remind
ourselves of the three theses distinguished in chapter 5, section I:
(i) Similarity- and comparison-judgments play a signicant, if not
essential, role in metaphorical interpretation.
230 Chapter 6

(ii) Metaphors and similes dier only by the occurrence in the latter of
the (supercial) `like'.26
(iii) Metaphors (i.e., utterances of declarative sentences containing at
least one metaphor) ``assert'' similarities.
As I argued in chapter 5, thesis (i) contains an important grain of truth
even if it is not the whole truth, but, in any case, it does not follow, contra
(iii), that metaphors assert similarities. The similarity judgments are part
of the context of the metaphorthe subject matter of the relevant pre-
suppositions about the relevant properties related to the word used meta-
phoricallyrather than its content. The corresponding constituent of the
content is simply a set of properties, not the similarity relation in virtue of
which those properties were singled out or identied.27
What of thesis (ii)? The surface grammatical form of a metaphor `a is
(an) F' is that of a one-place predication in which we say something
(expressed by the predicate) about something (the referent of the subject
term). It is not a two-place relation as the comparativist holds, and,
without compelling evidence, there is no reason to say that the underlying
logical form of the metaphor is dierent from its surface structure.
Nonetheless we can endorse (ii), so long as we understand it as saying not
that metaphors are (elliptic) similes, but that similes are metaphorical
predications and that they should be explained on the model of meta-
phorical predications. If the `like' of the simile is grammatically supercial,
similes, like metaphors, have (one-place) predicative structures in which
the predicate is interpreted metaphorically. As Goodman (1976) puts it:
``Instead of metaphor reducing to simile, simile reduces to metaphor; or
rather, the dierence between simile and metaphor is negligible'' (7778).
This also enables us to explain the observation (noted in ch. 5) made by
the psychologists Glucksberg and Keysar (1990) that it is always possible
to ``transform'' a nonliteral comparison or simile (e.g., `my kid's bedroom
is like a war zone') into a metaphor (`my kid's bedroom is a war zone'),
unlike literal comparisons (e.g., `my kid's bedroom is like the kitchen'),
which cannot be so transformed (*`my kid's bedroom is the kitchen'). The
reason is that the literal comparison statement really is a two-place rela-
tional statement; but the logical form of the simile is that of a one-place
predication. The `like' may be rhetorically or even grammatically eec-
tive, but it is not semantically signicant.
Why might one think that semantically metaphors are similes or com-
parison statements? Sophisticated comparativists like Fogelin begin with
the correct assumption that similarity plays a central role in working out
Metaphorical Character and Metaphorical Meaning 231

(some) metaphorical interpretations, from which they then infer (a) that
metaphors assert similarities as their propositional content, and (b) that
beneath the surface structure of the one-place metaphorical predication
there lies a relational statement that would be explicitly expressed by a
simile. Hence metaphors are ``elliptic similes.'' But this story fails for at
least two reasons. First, as Fogelin himself notes, it is not the literal
meaning of, or literal similarity expressed by, the simile that expresses the
gurative meaning, or gurative similarity, expressed by the metaphor;
rather it is the gurative meaning of the simile. Furthermore, the gura-
tive meanings of the metaphor and simile are also assumed (e.g., by
Fogelin) to be of one kind. But if so, the claim that metaphors are elliptic
similes does not explain the gurative meaning of the metaphor, say, in
terms of something nonmetaphorical or even in terms of something of a
dierent semantic kind. The analysis merely pushes the problem of ex-
planation one step back. Second, if the simile is the logically prior of the
two, and if the metaphor really expresses a (gurative) similarity relation,
some explanation must be given for the fact that the relational `like' is not
realized in the surface structure of the metaphor. Fogelin argues that the
metaphor is ``elliptic'' for the simile containing the overt `like', ``elliptic''
in the same sense in which (28) in reply to (27) in the following exchange
is elliptic:
(27) Are you coming?
(28) In a little while.
Here (28) is ``understood'' as and thus elliptical for
(29) I shall come in a little while.
However, this parallel is not sucient to make the metaphor/simile case.
For examples like (28), there is syntactic evidence for the presence of the
phrase `I shall come' in its underlying structure and evidence that it is
deleted in the surface structure. There is no analogous evidence that `like'
is really present in the underlying structure of a metaphor. Contrast
(30) Mary is coming now, and Jane in a little while
in which the deleted or copied verb `is coming' is understood, with
?(31) Mary is like the moon, Jane (is) like the stars, and Juliet is the sun.
Here there is no evidence that the last clause is ``understood''cannot be
interpreted exceptas containing an unrealized `like' and therefore con-
tains an elliptic `like', as does
232 Chapter 6

(32) Mary is like the moon, Jane like the stars, and Juliet the sun.
Likewise, there is nothing incoherent about saying
(33) Juliet is not like the sun, she is the sun
in which the predicate is interpreted metaphorically in both clauses.
However, on the elliptic simile view, this statement is elliptic for
(34) Juliet is not like the sun, she is like the sun
which is self-contradictory. On my view, that similes are metaphors, (33)
is not incoherent. Semantically, `is the sun' and `is like the sun' express the
same content (in the same contexts) but they dier rhetorically or prag-
matically. As many authors suggest, a simile is less direct and forceful
than its corresponding metaphor. The reason is not that the one is
``shorter'' or more concise than the other, but that `like' functions as a
hedge, or qualier, on the content. What is denied in the rst clause is the
qualication, not the content simpliciter, which in turn is armed with
emphasis in the second clause. Both clauses, however, are interpreted
metaphorically.28

VIII Is There One Natural Kind of Trope?

Let me now return to our opening question in the previous section. Are
metaphor, irony, simile, metonymy, and the rest of the tropes all members
of one natural kind of nonliteral interpretation or guration?29 If they
are, then if a given analysis works for one member of the kind, it ought to
work for alland failure to do so would ipso facto count as evidence
against the analysis, even for the case where it does prima facie apply.
The issue whether there is one natural class of gures is also important
for another reason. It motivates a further argument in the literature for
the view that metaphor is an illocutionary act or force, a kind of use that
falls under pragmatics rather than semantics. For if metaphor and, say,
irony are gures of one stripe, they both ought to be explained by one
kind of account. Now, whatever controversy surrounds the status of
metaphorical meaning, the ironic ``meaning'' of an utterance is surely not
a semantic meaning. If my wife comes home on the day on which I was
supposed to cook dinner, nds me working at the computer, the raw
chicken still in the freezer, and says, ``I could smell the aroma of roast
chicken blocks away,'' its semantic meaning is what the sentence carries
on its literal sleeve. It is not (solely) in virtue of our semantic knowledge
that we understand her utterance to mean that raw chicken does not make
Metaphorical Character and Metaphorical Meaning 233

a meal or that I have been utterly derelict in my duties. Nor are we even
tempted to posit an ironic meaning in the utterance in addition to the or-
dinary literal meanings of the words used. However the ironic interpreta-
tion is explained, it is a function not of the speaker's semantic knowledge
but of his use of his words to say more than they mean in virtue of their
force or illocutionary capacity. Assuming it is one of a kind with irony, a
similar story is then taken to apply to metaphor. Had my wife said in-
stead: ``You run on an Italian train schedule,'' it would be no part of our
semantic knowledge but a matter of the force or illocutionary capacity of
the utterance that would tell us that what she said is that I am unreliable
and always late. Thus, given this natural classication of the gures, it is
widely believed that their common account would be a theory of use or
pragmatics.
Whatever turns out to be the correct explanation of ironywhose full
account lies beyond the scope of this bookwhat primarily concerns us is
the step from irony to metaphor. Let's grant for the sake of argument that
irony is a kind of force (whether or not force is itself a kind of [speaker's]
meaning) or an illocution. It still does not follow that the same is true of
metaphor. In other wordsto put the claim back in the terms in which I
rst framed the questionthe account we have proposed for metaphor
does not generalize to all the other tropes because, I shall now argue, they
do not constitute a single natural kind. Instead I propose to divide the
received class of tropes into two main groups. The paradigm of the rst is
metaphor, of the second irony. The account I have proposed for meta-
phor is generalizable to the other tropes in its class, but it will not apply to
irony and its family.
Before turning to dierences between the two subclasses of tropes
which, for simplicity, I shall treat as the dierence between metaphor and
ironylet me mention two respects in which they are similar. It is also,
perhaps, because of these common features that others have tended to
lump them together. First, both are context-dependent. I have already
argued that this characterizes metaphor, but it may not be evident for
irony. Indeed several authors have claimed that irony, in contrast to
metaphor, is context-independent. Thus Ted Cohen writes that irony
``typically incorporates a function that leads from a given meaning to its
reverse or opposite'' (my italics).30 H. P. Grice proposes that, although
contextual inappropriateness (in light of the conversational maxims) may
indicate that an utterance is ironic, to determine the ironic interpretation
all we need to know is that the speaker
234 Chapter 6

must be trying to get across some proposition other than the one he purports to be
putting forward. This must be some obviously related proposition; the most
obviously related proposition is the contradictory of the one he purports to be
putting forward.31

And John Searle says:

Stated very crudely, the mechanism by which irony works is that the utterance, if
taken literally, is obviously inappropriate to the situation. Since it is grossly inap-
propriate, the hearer is compelled to reinterpret it in such a way as to render it
appropriate, and the most natural way to interpret it is as meaning the opposite of
its literal form.32

Neither Cohen, Grice, nor Searle acknowledges the role of context in

determination of the ironic interpretation (as opposed to its role in rec-
ognition that the utterance should be interpreted ironically). But as the
dierences between their accounts show, context must be brought in. In
particular, for Grice the ironic interpretation is the contradictorythat is,
the negationof the interpretation purported to be put forward; for
Searle and Cohen, it is the opposite or reversethat is, an interpretation
at the other end of the scale. Which of these is correct? Both, but each
only some of the timedepending on the context. Irony is a context-
sensitive function that yields a contrary from among a set of alternatives
themselves determined in the context. Very frequently, the relevant con-
trary is the opposite. For example, after having been beaten 800 in a
football game, one fan of the defeated team bitterly tells another, ``Well,
we really beat them, didn't we?'' However, in other cases, the ironic in-
terpretation may be closer to the contradictory, as Fogelin points out in
an example like the following. Suppose my son brings home an exam with
a grade of 80, for which he studied a bit, but not enough, because he
believed (and assured me) that he already knew the material cold. I com-
ment: ``Well, you really aced that one.'' The ironic meaning of my utter-
ance is neither its opposite, that he failed the examfor he did know
some of the material; nor is it simply the bare contradictory that he did
not completely master the material. For, as Fogelin indicates, the irony
would be inappropriate if the individual had missed just a single question
(e.g., received an A). Rather, it means something like: You have a long
way to go if you think you know that material and could do well on the
exam without studying. Yet, in another context, where my son did not
even try to learn any of the material, we can imagine the same utterance
meaning its opposite. Thus, as Fogelin also argues, irony, no less than
metaphor, seems to be a context-dependent gure.33
Metaphorical Character and Metaphorical Meaning 235

The second feature common to irony and metaphor, as Fogelin has

also argued, is that both involve mutually recognized intentions and
beliefsthat is, what I have been calling presuppositions. These pre-
suppositions enter into both recognition and interpretation of the two
gures. For both gures, the speaker intends his utterance to evoke, or
stimulate, in the respondent (who, in the case of irony, may or may not be
the target) a response, or adjustment to its literal meaning, which the
respondent, in turn, believes was intended by the speaker. Likewise, the
interpretation of the metaphor or irony must draw on beliefs, or inten-
tions, that the speaker believes his audience shares (or could share) and
that he believes his audience will believe he intends them to recognize.
Yet, despite these two features common to metaphor and irony, I now
want to turn to a dierence between them that points to their having dis-
tinct semantic statuses.
Consider the following complex gurative statement, that is, a statement
interpreted by more than one trope.34 Trying to decipher my clumsy,
awkward, sloppy, thick, and messy handwriting on a public document,
you say: ``What delicate lace work.'' In this context, the utterance is
interpreted both metaphorically and ironically. The same sentence might,
of course, be uttered on another occasion simply to say that someone's
handwriting shows care and carefulness, craft and training, a wonderful
attention to subtle calligraphic ourishes. In that context the utterance
would be a simple, hence nonironic, metaphor. And in yet a third context,
the same sentence might be uttered only ironically, say, as a comment on
some expensive curtains that your dog has just ripped to shreds. However,
let's imagine a context in which the speaker utters the sentence both
ironically and metaphorically. How should we describe the utterance? Is it
an ironic metaphor, or a metaphorical irony? The issue concerns the log-
ical order of interpretation. Do we rst interpret the utterance metaphor-
ically and only then determine its ironic interpretation? Or do we rst
determine the ironic interpretation of the literal vehicle used in the utter-
ancesay, nd the contextually appropriate contrary (i.e., a predicate) to
`lace work'and then determine the metaphorical interpretation of the
contrary expression? The question is not one of temporal order or of
actual psychological processing (although it may have implications for
these); the issue is rather whether one interpretation is conditioned on the
other.
Speakers I have polled unanimously seem to think that the ironic in-
terpretation is conditioned on the metaphorical interpretation of the
236 Chapter 6

utterance: First we determine the metaphor and only afterward the irony.
The reason for this ordering seems to be that the other ``irony rst'' order
is not only more dicult to compute; there is a conceptual diculty in
selecting the relevant ironic contrary to the literally interpreted term.
Would the contrary to ``delicate lace work'' in our earlier example
(according to the complex interpretation) be ``course rags'' or ``rough
sheepskin'' or ``a heavy shawl'' or ``sti polyester''? The diculty is that
we have no context-independent formula for deciding in a given case
whether the contrary is the contradictory or a polar opposite or some
contrary midway on the continuum from the mere contradictory to the
polar opposite. The element of the context that is most relevant to deter-
mine the appropriate contrary at this rst stage is information related
to the feature in terms of which the expression will then be interpreted
metaphorically at the second stage. So, to select an ironic contrary, it is
necessary to have some knowledge already of the metaphorical interpre-
tation of the expression.
This example is a relatively ``live'' metaphorical interpretation (in the
sense of ``live'' of chs. 1 and [as we'll see] 8, namely, an interpretation that
involves applying the character of the underlying metaphorical expression
to the presuppositions of its context to determine its metaphorical con-
tent). With dead, or dying, metaphors, the same problem does not arise. If
I say ``Shamir is a towering gure,'' intending the utterance to be inter-
preted both ironically and metaphorically, we might be able to interpret it
in either order. Either rst metaphoricallya man of impressive ability
and accomplishments, that is, of great stature, who commands great re-
spectand then ironicallyan unimpressive man of little ability and
accomplishments who commands little respect. Or rst ironicallya
diminutive gureand then metaphorically. Even here, however, there
may be subtle dierences in interpretation. And there is no guarantee that
there willalways or everbe a literal (ironic) contrary of the original
expression, which, under its subsequent metaphoric interpretation, will
express a feature contrary to the feature metaphorically expressed by the
original expression. However, the degree to which it is possible to reverse
the order of interpretation appears to vary with the degree to which the
metaphorical interpretation of the expression is dead, suggesting that
there may be no (or less) actual metaphorical interpretation taking place,
or that the interpretation may already be ``lexicalized.''
There is much here that is still not theoretically well understood.
However, we might draw a few tentative morals. First, and most impor-
tant, if irony and metaphor were straightforwardly two gures of one
Metaphorical Character and Metaphorical Meaning 237

natural kind, it is dicult to explain why there should be a prima facie

xed order of interpretation. One would expect, at the very least, much
more free ordering or more inversion of the order. That the order of in-
terpretation should always be metaphor rst and irony second suggests
that there is a dierence in kind between metaphor and irony as kinds of
interpretations. Furthermore, metaphor and irony do not appear to be
isolated gures in this respect. Similar results obtain with other gures,
suggesting that the received class of gures of speech can be divided into
the following two subclasses:
(M) Metaphor (I) Irony
Simile Meiosis
Synechdoche Hyperbole
Metonymy Understatement
Overstatement
I cannot conclusively demonstrate the hypothesis, but I shall tentatively
propose that whenever we have complex gurative interpretations, all M-
gures are interpreted prior to I-gures, that the latter are conditioned on
the former. There are ironic/hyperbolic, etc., interpretations or uses of
metaphors/similes, synechdoches, etc., but no metaphorical/similaic, syn-
echdochic interpretations or uses of ironies/hyperboles, etc.
The existence of a xed order of interpretation at most implies that
there is some dierence between the two classes of gures. It does not ex-
plain the dierence or even determine what the dierence is. As a second
conjecture, I want to propose that the (M)-gures are all character-istic
operations of interpretation, that is, operations on sentences (or their
tokens) that assign them nonconstant characters that, in turn, yield prop-
ositional contents in their contexts. The (I)-gures are operations on
propositional contents to yield (dierent) propositional contents. So, for
the other (M)-type gures, we might also posit operators like `Mthat' that
take arbitrary context-independent expressions and yield metaphorical-
like expressions whose nonconstant characters are sensitive to a cor-
responding contextual parametera context set of relevant presupposi-
tions. (Where there is no (M)-type interpretation, the literal meaning of
the sentence directly generates the content of the sentence.) The contents
generated by the characters of the various M-type expressions, in turn,
can be interpreted by an (I)-gure, say, ironically. Hence (M)-type gures
are semantic interpretations, interpretations determined by the semantic
structure of the language; whereas (I)-type gures are postsemantic, that
238 Chapter 6

is, uses of the semantic interpretations of sentences, namely, propositions,

to yield further propositions.35 This dierence between the two classes of
tropes also explains our intuition that sentences meant ironically none-
theless are understood literally, whereas metaphors are not understood
literally even if, in some sense, they still ``have'' their literal meaning. In
any case, even apart from this nal conjecture, it should be clear there is
good reason not to draw an inference from the pragmatic status of irony
to a similar conclusion for metaphorwhich answers the objection with
which we began. Despite received opinion, there may be no single natural
kind of gures of speech.

IX Three Semantic Theories of Metaphor: A Comparison

In recent years, semantic theories of metaphor have not been nearly as

fashionable as pragmatic theoriesand (as I argued in ch. 1) they are also
not the natural rst candidates to which we would turn for an explanation
of metaphor. But mine is also not the rst or the only semantic theory in
the eld.36 To emphasize its distinctive features, I'll conclude this chapter
by contrasting it with three other semantic theories in the literature and,
in the next section, by replying to several potential objections.
The rst kind of semantic theory of metaphor, or class of theories, had
its greatest appeal and strongest inuence in the early 1970s. To explain
both metaphorical recognition and interpretation in one fell semantic
swoop, theories in this vein diagnosed the ``tension'' frequently identied
with metaphor in terms of the then almost universally accepted claim that
a metaphor is, taken literally, semantically anomalous or grammatically
deviant.37 That is, its constituent expressions violate various semantic (or
perhaps syntactic) co-occurrence conditions if the utterance is taken liter-
ally. This fault of the sentence uttered, it was argued, excludes its literal
interpretation, from which these authors also concluded that the utterance
is ipso facto identied as a metaphor; and through the same violation of
co-occurrence conditions the metaphorical interpretation is generated.
According to some authors, a new metaphorical sense emerges from sec-
ondary connotations to which the interpreter shifts when the literal
meanings fail; according to others, it is the result of cancellation-,
weighting-, and transference-operations performed on component fea-
tures of the lexical entries of the words. On all these accounts, however,
metaphorical meaning is a reinterpretation of language that cannot be
understood in its primary, privileged literal sense. The metaphorical in-
terpretation is nevertheless semantic, for it is a function of violations of
Metaphorical Character and Metaphorical Meaning 239

semantic conditions, conditions the speaker knows as part of his linguistic

competence. Metaphorical interpretation falls within the grammar, or se-
mantics, of a language because all we need to know in order to explain
how it works are the linguistic or semantic rules that it violates.
As I noted back in chapter 1, the fatal weakness of all these theories is
their ruling assumption that all, or most, or at least typical metaphors are
grammatically deviant, semantically anomalous, or just plain false under
their literal interpretation. In the mid-1970s this dogma was decisively
challenged. In Ted Cohen's apt phrase, there are many ``twice-true'' met-
aphors, utterances that are perfectly ne and equally true in the very same
context whether they are interpreted literally or metaphorically. More
generally, the import of the critique is that there need be absolutely noth-
ingsyntactically, semantically, or even pragmaticallydeviant or
irregular about the literal meaning of a sentence used or interpreted met-
aphorically. (Of course, it would not help a semantic account of metaphor
even if metaphors were pragmatically odd.) Hence neither the recognition
nor interpretation of a metaphor can be explained by way of its literal
impossibility, let alone its literal semantic impossibility.
Unlike these accounts, mine does not assume that metaphors are
semantically or grammatically (or, for that matter, pragmatically) deviant
or anomalous under their literal interpretations, or that they should be
interpreted in terms of violations of semantic (or indeed any) rules. Fur-
thermore, my theory is diametrically opposed to the view that metaphors
are reinterpretations of strings that cannot be taken according to another
prior, privileged sense. On the contrary, the metaphorical expressions of
the form `Mthat[F]' that underlie metaphorically interpreted utterances
are fully grammatical; they are generated by the grammar as a subset of
the set of expressions of nonconstant character, indeed in parallel and
simultaneous with the nonmetaphorical characters in the metaphor
character set under which the utterance can be typed (ch. 4, sec. IV).
Knowledge of the characters of metaphorical expressions falls squarely
within the domain of a grammar, or our linguistic competence, like that
of the characters of context-dependent but literally interpreted indexical
expressions. To be sure, the propositional content of the utterance of a
metaphor requires appeal to an extralinguistic parameter (i.e., its presup-
position set), but so does the content of `I' (its agent) or `here' (its time) in
its context. Thus knowledge of these contents, determined by their re-
spective extralinguistic parameters given their nonconstant characters, is
also extralinguistic or not purely semantic. But to be extragrammatical in
this way is not to be ungrammatical.
240 Chapter 6

In claiming that metaphorical expressions are grammatical, I do not

merely mean to say that it is possible to accommodate metaphor within a
grammar by making revisions or additions compatible with standing
syntactic or semantic rules. On my account, metaphors require an abso-
lute minimum of semantic apparatus beyond what is already in the
grammar, in addition to what is independently necessary for the semantic
interpretation of nonmetaphorical language. Because `Mthat' is simply a
variant of `Dthat,' the main apparatus is already available for the se-
mantics of demonstratives. All that needs to be altered in the grammar is
the addition of the one operator `Mthat' to the lexicon, together with a
rule that (optionally) spells it out either as the word `metaphorically' or
with appropriate stress. On this count, my theory diers from previous
approaches that attempted to incorporate metaphor within grammar by
rewriting or adding substantively new grammatical rules. Unfortunately,
in making the grammar descriptively adequate for metaphor, these
theories in turn weakened its explanatory power.38 A distinct virtue of my
account is that it leaves the grammar and semantics in place; because it
adds no signicantly new rules, there is no cost to the explanatory power
of the grammar.
The second kind of semantic theory of metaphor begins from the
assumptions that the literal is semantic and that the metaphorical depends
on the literal. It assigns each expression in the language a set of compo-
nent features (``selectional features'' or ``semantical hypotheses'') that
constitute its lexical description or entry, which it in turn identies in toto
with the literal meaning of the word. The various metaphorical inter-
pretations of the word are then derived by dropping or deleting one or
another feature, thereby highlighting or weighting the remaining ones.
Thus ``the metaphorical meanings of a word or phrase . . . are all con-
tained, as it were, within its literal meaning or meanings. They are
reached by removing any restrictions in relation to certain variables from
the appropriate section or sections of its semantical hypothesis.''39 That
is, a metaphorical interpretation is the result of cancellation of features
given in the literal, or lexical, meaning of the expression.40 And because a
speaker presumably knows the literal, or lexical, meaning of a word as
part of his linguistic and semantic competence, it follows ipso facto that
he knows its metaphorical meaning as part of his semantic competence as
well.
One attraction of this approach is its straightforward explication of the
way in which the metaphorical ``depends'' on the literal, in terms of its
idea that the one is contained in the other. But this claim is also the major
Metaphorical Character and Metaphorical Meaning 241

shortcoming of this account. The diculty is that we must both specify

the literal or lexical entry of a word broadly enough to include the wide,
variegated range of properties that it can express metaphorically and yet
remain faithful to the idea that these properties are known by a speaker as
part of her semantic competence. It is impossible to satisfy both of these
descriptive and explanatory desiderata simultaneously. One need not
assume a sharp analytic-synthetic distinction to make this point. Even if
one takes common knowledge to be part of the ``dictionary meaning'' of a
wordincluding false but stereotypical features (such as Searle's example
of the stereotypical features of being mean and prone to violence asso-
ciated with the name `gorilla')what many metaphors express in context
are not antecedently accepted beliefs but properties noticed on and
restricted to that occasion.41 Indeed the problem is exacerbated insofar as
we make cancellation of features depend on relations between the features
of the metaphorical vehicle and features of other expressions in the string.
Consider the dierent metaphorical interpretations of `is a sh' in
(35) Grace Kelly was a sh
(36) Richard Nixon was a sh
say, `is at home in water' in (35) and `is slippery and hard to catch' in
(36). These two strings raise two problems. First, the two dierent meta-
phorical interpretations of `is a sh', as I have stated them, are themselves
expressed (at least arguably so) by metaphors.42 Suppose, moreover, that
it is only possible to express the metaphorical interpretations using meta-
phors. I'll argue in chapter 7, section VI, that this is not problematic for
my account but, rather, typical and indeed necessary in order to capture
the cognitively signicant ``perspective'' on its content contributed by the
character of the metaphor. However, for a lexicalist semantic theory like
the one under consideration, it poses a serious problem if the content of a
given metaphor can only be adequately expressed by employing other
metaphors. If the lexical entries must themselves consist in features that
can only be expressed by metaphors, then the theory must surrender its
stated motivation to explicate how the metaphorical ``depends'' on the
literal in terms of (feature) ``containment'' in a lexical entry.
The second problem raised by (35)(36), to return to my main line of
argument, is that the dierence between their interpretations is in part
prima facie dependent on presuppositions associated with the proper
names for the respective individual subjects of the sentences. But proper
names cannot be lexically represented by the necessary features to mark
these dierences without making every presupposed property of an indi-
242 Chapter 6

vidual a lexical feature of her name. Indeed, if we try to build all these
features into lexical entries, it soon becomes clear that the notion of lex-
ical feature is simply doing the work of encoding what presuppositions are
made in a given context, including many that are extralinguistic. The
theory that results is hardly semantic.
The third semantic theory I will discuss is Eva F. Kittay's ``perspectival
theory,'' which, using the lexical semantics of semantic eld theory, takes
the meaning of a term to be a function of its relations of anity and
contrast to other terms in its eld.43 According to Kittay, the interpreta-
tion of a metaphor (like `Juliet is the sun') involves the transference, or
mapping, of the semantic eld associated with the vehicle of the metaphor
(e.g., the eld of terms for celestial bodies such as `the sun') onto the do-
main of the semantic eld associated with the topic of the metaphor (e.g.,
the domain of women or humans associated with `Juliet'). In mapping the
former onto the latter, the speaker reconceives, or restructures, the latter
in terms of the intralinguistic relations that hold within the system of
expressions in the former. With this restructuring of the topic domain,
there emerges what Kittay calls the second-order interpretation that con-
stitutes the meaning of a metaphor.
There are several similarities between Kittay's theory and mine. Both of
us emphasize the motif that metaphorical interpretation involves system-
atic families or sets of expressions. The central role I assign to networks in
the interpretation of a metaphor parallels the role Kittay assigns to se-
mantic elds, although she is more inspired by Black, I by Goodman. As I
also mentioned in chapter 5, section III, the semantic eld theoretic no-
tion of paradigmatic relations that hold among members of a contrast set
corresponds to that of the exemplication relations that hold for members
of the sample scheme to which the referent of the literal vehicle of the
metaphor belongs. The syntagmatic relations of semantic eld theory can
also be analyzed more generally as the network of expressions that are
candidates to ll the argument places marked by the thematic relations
that underlie the expression interpreted metaphorically. Kittay's detailed
analyses of a number of complex metaphors oer good illustrations of the
rich interpretations that can be captured when, and only when, we pay
attention to the systems of expressions, and their relations, in which indi-
vidual metaphors function.
Yet there are also several deeper dierences between Kittay's theory
and mine that reect our dierent conceptions of semantics and context,
two notions that are intimately related in both of our accounts. On the
one hand, I have argued that metaphorical interpretation is highly sensi-
Metaphorical Character and Metaphorical Meaning 243

tive to contextual presuppositions, some of which may be known by the

participants as part of their lexical knowledge but most of which are
extralinguistic. To avoid drawing a sharp analytic-synthetic-like dis-
tinction, I have treated all of these together under the rubric of pre-
suppositions, hence (for all theoretical purposes) as a nonlinguistic
parameter of semantic interpretation. On the other hand, I have also
argued that speakers possess semantic competence specic to their ability
to interpret metaphors that is central to their knowledge of language (as
opposed to empirical beliefs). This is their knowledge of metaphorical
character, which constrains their interpretations. Insofar as metaphorical
character is an instance of character in general, which is central to
speakers' knowledge of language, I have argued that this account has
genuine explanatory force. It subsumes metaphorical interpretations un-
der the interpretation of context-dependent expressions (demonstratives
and indexicals) tout cour. The main dierence between Kittay's theory
and mine is that she (apparently by design) does not draw any sharp dis-
tinction between speakers' knowledge of language or semantic compe-
tence, on the one hand, and their empirical, extralinguistic beliefs and the
kinds of skills that enter into their use of language, on the otherwhat I
call ``context.''44 Hence her explanatory aim must be rather dierent than
mine. In giving what she calls a ``semantic theory of metaphor'' her point
does not seem to be to give a theory that attempts to isolate what speakers
know specically as part of their knowledge of language or semantics.
Therefore, despite the descriptive riches of her analyses, there is no spe-
cic explanatory force to her claim to have given a semantic theory of
metaphor. Let me illustrate this dierence between our theories by refer-
ence to two of Kittay's explanations and certain problems they raise.
Like the rst kind of semantic theory we discussed, Kittay claims that a
metaphor ``breaks certain rules of language, rules governing the literal
and conventional senses of terms'' (ibid., 24). She employs this assump-
tion of literal incongruity rst to explain how we identify a metaphor, but
it is also deeply entrenched in her conception of metaphorical interpreta-
tion insofar as the literal incongruity reects the conceptual dierences
that demarcate the semantic elds of the vehicle and topic, the dierences
in terms of which she characterizes the tension distinctive of metaphor.
This assumption of literal incongruity is already one issue that sharply
distinguishes Kittay's theory and mine, but let me grant it for the moment
in order to focus on a deeper dierence between us that emerges from her
replies to the various counterexamples that have been raised against the
assumption.45
244 Chapter 6

To defend her claim about literal incongruity, Kittay argues that the
purported counterexamples mistake the proper unit that is literally in-
congruous. Sometimes it is an immediate constituent phrase (e.g., Eliot's
`a slum of bloom'), at other times the sentence uttered, but, where neither
of these is prima facie incongruous, the relevant unit is the utterance in its
situational context. In these cases, however, Kittay is not satised simply
to uncover some kind of incongruity, such as pragmatic or conversational
oddity. She tries to demonstrate further that the presence of one of these
latter kinds of oddity is simply symptomatic of the presence of genuine
semantic incongruity. To construct her case, she proposes the Expres-
sibility Principle: All salient elements of a situational context can be
expressed in linguistic terms. So, if the context of a metaphor does not
consist in a linguistic text (e.g., an explicit verbal discourse) containing a
semantic incongruity, Kittay claims that the salient features of the non-
linguistic context can always ``be rendered linguistically as an utterance of
a level of complexity higher than that of the given expression'' (62). This
linguistic context, she then conjectures, will contain a violation of the se-
mantic combination rules that govern rst-order literal language, hence a
semantic incongruity. As a consequence, she concludes, this ``places the
identication of metaphors squarely within the province of semantics''
(75).
This conclusion is symptomatic of the problems with Kittay's concep-
tion of semantics and context. Consider the example:
(37) The rock is becoming brittle with age
where the subject description is used metaphorically in a context to refer
to an aging professor, ``accompanied perhaps by a gesture, for example, a
nod in the direction of the professor'' (71).46 The sentence (37) violates no
semantic or linguistic rules and contains no incongruity but, Kittay
claims, the utterance in its situational context does. Therefore, she pro-
poses that we ``render'' the gesture, as part of the situational context,
``linguistically,'' and thereby provide an incongruous frame for the sen-
tence (37) as metaphorical focus. For example, she imagines (37) in the
``context'' of (38):
(38) He responds to his students' questions with none of his former
subtlety.
Here Kittay reasons that the pronoun in (38) must be anaphorically cor-
eferential with `the rock' if the two utterances are to cohere conversa-
tionally. Therefore, we ought to be able to substitute the antecedent for
Metaphorical Character and Metaphorical Meaning 245

the pronoun in (38). However, this immediately creates a semantic viola-

tion since `the rock' is [-animate] and the verb `responds' requires a sub-
ject that is [+animate]. To make the incongruity explicit, she suggests that
we construct a ``conversion sentence'' that states the identities and equiv-
alencies of terms:
(39) The rock is the professor.
This is a possible description of an interpreter's reasoning. But as an
explanation of how we identify an utterance of (37) as a metaphor, the
utterance in a situation (37)(38) and the conversion sentence (39) fail to
show that it is by means of exclusively semantic knowledge. First,
although it is possible for the pronoun in (38) to corefer to the referent of
`the rock' in the previous sentence, it is not linguistically, or semantically,
obligatory that it do so.47 The pronoun might equally well be functioning
as a demonstrative, demonstrating the individual rendered salient in the
context (e.g., by the independent metaphorical reference to the person in
the previous utterance). Hence to link the two we already need to know
that `the rock' refers to the professor in its context, that is, we need to
know the very information that is expressed by the conversion sentence
(39). However, knowing (39) or the fact that in the context `the rock'
refers to the professor requires nonsemantic empirical knowledge. Hence
there is no conceptual or linguistic connection between (39) and (37), only
an empirical one, and the fact that (39) involves an incongruity does
nothing to explain away (37) as a congruous counterexample to the in-
congruity thesis. Second, Kittay argues that when we sense an oddity
we construct a linguistic context like (37)(38) that violates a semantic
rule. But it is clearly only because we already believe that (37) is meta-
phorical that we recover the context (38). Therefore, we cannot appeal to
the violation involved in (37)(38) to explain why we take (37) to be
metaphorical.
In short, the problem with this explanation is that even granting Kittay
the assumption of literal incongruity, one cannot assimilate the role of the
extralinguistic context of an utterance into semantics as she does without
robbing the attribution of semantic knowledge of its explanatory
power.48 On my view, recognition of a metaphorthe knowledge that an
utterance is to be interpreted metaphoricallyis a matter of recognizing
that an utterance is a token of one rather than another type, where the
types are individuated (in part) by their respective meanings (or charac-
ters). Our semantic knowledge is knowledge exclusively of the types with
their characters; what enables us to subsume utterances as tokens under
246 Chapter 6

types introduces nonsemantic knowledge, for example, about the purpose

of the utterance and other features of the extralinguistic use of the sen-
tence. This task falls to a theory of speech or pragmatics.
There is a similar problem for Kittay's use of semantic elds to explain
metaphorical interpretation. Although there is considerable controversy
over the exact denition of a semantic eld in all its complexity, the basic
idea is that it is a set of contrast sets and possibly a permutation relation
dened over the members of the sets.49 A (basic) contrast set hL: E1 . . .
En i is a sequence containing a linguistic expression L that serves as a
covering term for the set and a set of expressions E1 . . . En such that the
extension of each Ei is a subset of the extension of L and the extensions of
each Ei are disjoint. Examples of (basic) contrast sets are: hcolors: red,
green, blue, yellow, orange, brown, purple, black, whitei and hdays:
Sunday, Monday, Tuesday, . . . , Saturdayi.
As long as we limit ourselves to (basic) contrast sets like the two
examples just mentioned, it is plausible to hold that knowledge of the
semantic eld to which a given expression belongs is part of a speaker's
lexical and, therefore, linguistic knowledge. However, as we already saw
in connection with the second kind of semantic theory, as soon as one
tries to extend the idea even to relatively simple metaphors, it becomes
less and less plausible that the networks in which the terms are function-
ing could possibly be known by speakers in virtue of their linguistic or
semantic competence. Again, consider one of Kittay's own examples.
(40) The seal dragged himself out of the oce.
Let's grant Kittay the assumption that in its (verbally explicitated) con-
text an utterance of (40) contains a semantic incongruity, say, between the
vehicle `seal' and topic `human beings'. However, as Kittay herself shows,
it is possible for utterances of (40) to be interpreted metaphorically with
two distinct contents. On the rst interpretation, let `seal' express the
property of being a performing creature who knows all the tricks and is
constantly put on display, subject to its captor's purposes. On the second
interpretation, `seal' metaphorically expresses a number of features con-
cerning physical appearance and manner: wearing a black shiny suit,
having dark slicked-down hair, a waddling walk. Further, let's grant that
utterances of (40) in their respective contexts could have either of these
two metaphorical interpretations. In virtue of what kind(s) of knowledge
will the speaker know the content of either of these distinct metaphorical
interpretations of (40)? Kittay proposes that the contexts in which the two
interpretations result would include the dierent networks of expressions
Metaphorical Character and Metaphorical Meaning 247

to which `seal' respectively belongs. On the rst interpretation, the net-

work might include seals as captive, performing animals exploited by
their owners for the latters' amusement and benet, as opposed to free,
noncaptive, unexploited animals who are masters of themselves. On the
second interpretation, the network includes the features of seals' physical
appearances, both in contrast to each other and to those of other crea-
tures, features like a lion's mane or an elephant's wrinkled, saggy skin.
As a description of how the interpretations are determined, this story is
entirely plausible. The question for our purposes is whether knowledge of
these networks or sets of expressions that determine these interpretations
is semantic. If the semantic eld of an expression is determined by its
lexical/linguistic/conventional meaning, `seal' in (40), even as it is used to
yield these divergent interpretations, ought to belong on both occasions to
one semantic eld. But relative to one constant semantic eld, we cannot
explain the dierent metaphorical interpretations of (40). In reply to this
issue, Kittay writes that, as these examples illustrate, ``most terms belong
to a large number of semantic elds. Which semantic elds come into play
in the interpretation of the metaphor depends on what [literal/conven-
tional] interpretation we have chosen for the term'' (165), and this, in
turn, depends on the literal/conventional senses of the other terms in the
discourse. Yet, for an example of this process of selection, all she says
is that in the case of `seal' ``we tend to eliminate the sense of `seal' which is
an `inanimate stamp' '' (165166). Yes; but that sense of `seal' diers in its
syntactic and semantic feature components from the sense of `seal' in (40).
An explanation of dierent interpretations of (homonymous) tokens that
in fact belong to dierent types with dierent syntactic and semantic fea-
tures hardly explains our two metaphorical interpretations, whose literal
vehicles have identical syntactic and semantic features. Perhaps to ll in
this lacuna, Kittay next argues that the semantic eld of an occurrence of
the term is only ``fully specied when we obtain the most complete
[literal/conventional] interpretation of the utterance in which the term's
[literal/conventional] sense is a constituent'' (166). By a ``full specica-
tion'' she seems to mean: making verbally explicit the context of the utter-
ance according to her Expressibility Principle, an explicitation that will
make explicit the term's contrast set. How interpreters accomplish this,
she concludes, is the ``dicult question of how a context selects out rele-
vant semantic elds'' (ibid.). About this, unfortunately, she has nothing
more to say other than that ``work has yet to be done in this area'' (ibid.),
although she herself concedes that the relevant considerations will be in
large measure pragmatic. In short, if we want to appeal to dierent con-
248 Chapter 6

trast sets or elds to explain the dierent metaphorical interpretations of

tokens of one expression type, we must introduce nonsemantic, extra-
linguistic informationsuch as the dierent exemplication relations of
seals relative to dierent sample/exemplication schemata to which they
belong. Again, the moral is that there is no explanatory force to Kittay's
notion of a semantic eld.50 Of course, a speaker may know as part of his
semantic lexical competence that a term like `seal' belongs to a contrast
set with other captive and noncaptive animals. But knowledge of the
content of the contrast, the particular features in virtue of which the
expressions stand in contrast, goes far beyond what speakers know in
virtue of their semantic or lexical competence.
If we want to capture a body of semantic knowledge that underlies
metaphorical interpretation, we cannot look, then, only to the networks
in which the metaphorical vehicle and topic function (even if knowledge
of some of the relevant networks may be part of our linguistic compe-
tence). Making the contrast sets verbally explicit and then arguing that it
is our knowledge of the intralinguistic relations between the explicitated
terms that determines the interpretation simply disguises the fact that we
know what to make explicit in virtue of extralinguistic knowledge and
beliefsthat is, contextual presuppositions.51 On my theory, on the other
hand, what is known by the speaker in virtue of his semantic competence
is not the schemata or networks but the character of the metaphor, the
appropriate function, with its constraints, from contextual presupposi-
tions to contents. Finally, note that Kittay also employs a function that
maps the rst-order literal/conventional interpretation onto the second-
order metaphorical interpretation. However, she also emphasizes that this
function is just one member of a whole set of functions that yield second-
order interpretations, including indirect speech acts, irony, and non-
gurative utterances with two illocutionary forces. The latter are, quite
clearly even on Kittay's account, nonsemantic, pragmatic interpretations.
She gives us no reason to believe that the function specic to metaphorical
interpretation is any more semantic.52

X Objections and Replies

Objection: Our ultimate goal is to account for the acceptability or success

of metaphorical utterances. Semantic theories like yours promise to do
this by tying acceptability or success to truth (and unacceptability or
failure to falsity). But this move presupposes that we have at least as good
a grasp, if not a better grasp, of the truth and falsity of metaphors
Metaphorical Character and Metaphorical Meaning 249

hence of their truth-conditions or contentas we have of their success

and failure. Yet, we all know how dicult it is to know with certainty that
a metaphor means one thing rather than another. Metaphors can often be
interpreted in multiple, sometimes incompatible, ways (``depending on
context''), and metaphorical interpretation seems endless at least in the
sense that we often don't know when to stop. As Donald Davidson puts
it, it is an error to think that metaphors have xed, determinate truth-
conditional contents, for if they did, why is it ``often so hard to decide . . .
exactly what the content [of a metaphor] is supposed to be'' (WMM,
262)? On the other hand, if metaphors do not have determinate contents
or truth-conditions, we cannot condition their success (or failure) on well-
dened criteria of satisfaction of those conditions. Invariably we must
have recourse to pragmatic, nonsemantic desiderata to explain their suc-
cess and failure. But in that case truth-conditional semantics does no
explanatory work. Hence there is no need to advert to semantics or
truth-values for metaphors, indeed good reason (viz., Ockham's razor)
not to.53
Reply: The objection charges that without a rm knowledge of meta-
phorical contents or their truth-values, we cannot hope to explain why
metaphors succeed by attributing semantic knowledge to their speakers.
To begin with, it is important to keep in mind that it was never claimed
that a speaker's semantic knowledge of metaphor is sucient for him to
interpret a given utterance of a metaphor, hence, to account for its con-
tent and its actual truth-value. For that purpose, knowledge of the con-
text, or presupposition set, is also necessary. So, if we don't know, or ``it's
hard to decide,'' the truth-value or truth-conditions of the metaphor, the
diculty may be due to at least two possible reasons. One possibility is
that metaphors do not have ``determinate'' interpretations and therefore
really do lie beyond the scope of semantics. Or it might be that we simply
do not know the relevant aspects of the context that must be known to
interpret the metaphor, that is, we lack extralinguistic knowledge. Before
rejecting the semantics, we need to discount the latter possibility.
Of course, even if we shouldn't conclude from the fact that we don't
know the content or truth-value of a metaphor that it has none, one may
still object that it is not obvious what advantages accrue from the as-
sumption. My proposal is twofold. First, by assuming that metaphors are
truth-valued and have contents, we can articulate more clearly what we
don't know when we don't know how to judge at least one kind of success
of a metaphor, the kind of success that turns on the metaphor having a
determinate metaphorical interpretation. One such factor that might
250 Chapter 6

enable us to explain what makes it so ``hard to decide'' on the content of a

metaphor is its context set of presuppositions, the extralinguistic parame-
ter of its interpretation that we can separate out only by distinguishing it
from the speaker's semantic knowledge (of character). For example,
where the metaphor originated in a historically remote or culturally for-
eign context, what can make it so hard to decide on its content is that we
often do not know the author's presuppositions unless a good literary
historian tells us or an anthropologist discovers them. Another reason it
can be dicult is that there is more than one possible set of pre-
suppositions among which we have no good reason to choose one rather
than another as the context for the metaphor. Furthermore, even when we
can begin to spell out presuppositions, we often don't know when to stop:
how to delimit the presuppositions and thereby delimit the content. Be-
cause of the interpretive power of metaphorical character, the longer we
dwell on many metaphors, the more, and the more novel, presuppositions
we frequently make.54 (Again, for rich, vital metaphors, there may be not
just one but alternative, even mutually incompatible, kinds of potential
presuppositions.)
How, then, do we know when to stop to settle on a xed metaphorical
content? What determines what does or does not belong to the context of
a given metaphor so we can be condent, at least as an idealized standard,
that there are some metaphors with determinate truth-value-bearing con-
tent, relative to which we can then explain why the ideal is not realized in
other cases?
In a word, we do: We x the contexts of our metaphors. `Juliet is the
sun' out of context could mean that she is to be worshipped or that she
can burn up her admirers or that she follows regular predictable motions
or. . . . But x the context of the utterancethat is, the specic context set
of presuppositionsand ipso facto we x its content in that context.55
The fact that the metaphor could mean this in one context and mean that
in another does not show that within a given context it has more than one
of these interpretations. And to the further charge that metaphorical in-
terpretation even within a context is endless and indeterminate, my reply
is: only if we let it be so. It is up to us to impose a closure. To be sure,
given a xed context set of presuppositions, it is always possible for us to
add new ones. But from the point of view of the semantics, each such
addition constitutes a dierent context relative to which the metaphor has
a distinct interpretation.
The second half of my proposal centers on what we should and what
we should not expect of a semantics of metaphor. The semantics, on our
Metaphorical Character and Metaphorical Meaning 251

account, does not furnish a unique method or procedure to generate the

presuppositions, an algorithmsomething like the traditional grounds of
metaphorthat would enable us to produce a (literal) explication or ex-
plicit statement of the content of a metaphor in terms of the presup-
positions. For that we need a pragmatics or theory of context. But this
should not count against a semantic theory of metaphor any more than
the fact that we cannot individuate the contents of `here' and `now' with-
out theories of locations and times should count against a semantic theory
of the indexicals. The point of the semantics is not to generate actual
interpretations, let alone give rules or recipes that will make it easier for
us to interpret metaphors. The semantics only explains constraints on
possible metaphorical interpretations by spelling out their form or struc-
ture relative to certain parameters. This structure of metaphorical inter-
pretationmetaphorical characteris just as rich (and as impoverished)
as the structure of interpretation of indexicals; so I argued earlier in this
chapter (secs. I, III, and IV). By the same token, if the semantic structure
of a metaphor underdetermines its interpretationand allows for an
interval of alternative interpretations rather than a unique content (for
some metaphors)this kind of vagueness or indeterminateness should
not count against the truth-valuedness of metaphors any more than it
should count against the truth-valuedness of literal utterances containing
indexicals or demonstratives that their ``neighborhoods'' are also some-
times vague or indeterminate. The constraints for the two classes of
expressions stand (or fall) together.
Objection: If the reason we sometimes do not know a determinate con-
tent for a metaphor is that we do not know its contextual presuppositions,
it is no surprise that once we know the presuppositions, we know the
contentthey are one and the same. Hence your semantics tells us noth-
ing more about the interpretation of a metaphor than we already know
when we know the context set of presuppositions. Furthermore, isn't the
reason we don't know the context set of presuppositions because we do
not know, independently of knowing the content, which presuppositions
we must hold? Without an independent account of how to identify the
extralinguistic presuppositions, the semantics is trivial.
Reply: You are right insofar as a theory of our semantic competence in
metaphor does not itself generate the interpretation of a particular meta-
phor in a context. For that we need extralinguistic knowledge. But if the
semantics tells us that the (content of an) interpretation of a metaphor
depends on its actual context set of presuppositions, and on nothing but
that, it is hardly fair to charge that now that we know the context set of
252 Chapter 6

presuppositions, and thereby the content of the metaphor, what more does
the semantics tell us.
There is also a second point to the objection. I have emphasized that
the semantics must be complemented by a pragmaticsby a theory of
how we apply our semantic competence to particular contexts, that is, sets
of presuppositionsand I have rehearsed various stories (in ch. 5) about
the kinds of desiderata (e.g., the various networks to which the vehicle
belongs) that ground the presuppositions for some kinds of metaphors.
So, to the extent to which we can describe these factors that yield the
presuppositions, we can give an independent characterization of the met-
aphorically relevant or m-associated presuppositions, as the objection
demands. However, as I have also said, there is no sure-re criterion to
pick out a unique set of relevant presuppositions for a given metaphor.
Indeed this is one way I suggested that our semantic competence that
consists in knowledge of the constraints that govern character diers from
the kind of extralinguistic knowledge of the context that constitutes our
pragmatic skills.
It is also true, as the objection says, that we generally don't know which
presuppositions we must hold independently of knowing the content of
the metaphor. In some contexts we do have independent means of know-
ing the m-associated features presupposed to be associated with the literal
vehicle. But the presuppositions that must be held are those that are
``required'' (or at least are sucient, even if not uniquely so [see ch. 4, sec.
III]) for the interpretation of the metaphor, just as the pragmatic pre-
suppositions of utterances generally are those required for their appro-
priateness. Indeed, as I suggested in chapter 4, we accommodate our
contexts to the interpretation of our metaphorical utterances. This, how-
ever, is not an objectionable way in which metaphors are exceptional, but
a general characteristic of the interaction between contextual presupposi-
tions and utterances.
Objection: According to your theory, the interpreter of a metaphori-
cally used expression F is said to know, as part of his semantic compe-
tence, the character of the metaphorical expression `Mthat[F]', that is, a
rule from the context set of presuppositions to the content of the expres-
sion in that context. But other than readers of this monograph, surely no
speaker of English or of any other natural language has ever heard of
`Mthat'. Therefore, no such speaker has knowledge of `Mthat' or of met-
aphorical character.
Reply: It is helpful to distinguish two issues here. The rst issue is
whether it is theoretically fruitful to attribute cognitive states to speakers
in order to explain their ability to interpret metaphors. I have tried to
Metaphorical Character and Metaphorical Meaning 253

argue that it is both productive and necessary to posit such cognitive

states, insofar as accounts that purport to account directly for our use of
metaphors in terms of speech act, conversational, or verbal abilities are
inadequate; see chapter 2 and earlier in this chapter. But whether or not
you agree with those arguments, it should be understood that being in
such a cognitive state or having such knowledge does not require that
speakers or interpreters also be able to verbally articulate or explicitly
state the content of those states, such as the rule of character for `Mthat'
or indeed any semantic principle that is the content of the state they are in
when they are semantically competent in a language. Nobody but readers
of this book, admittedly, are likely to be able to do the latter. That,
however, hardly counts against my theory as the best explanation of the
cognitive state of interpreters inferable from the data at hand. Further-
more, on this score, knowledge of the rule of character for metaphorical
expressions is no dierent from knowledge of any rule or principle of
grammar (e.g., of the sort spelled out in generative grammar), none of
which can be assumed to be explicitly known or statable by those bearing
the cognitive states of which they are the content.
The second issue is whether the cognitive state that consists in grasping
the character of Mthat-expressions, or any principle of grammar or se-
mantic competence, should be described as knowledge, that is, whether it
is appropriate to say that interpreters know the characters of metaphors.
If you do think that ``knowledge'' is an appropriate term for this state, the
knowledgewhich, again, is of the same kind as linguists attribute to
speakers when they say that they know a language or its grammarmust
be allowed to be tacit, implicit, or unconscious. If you don't think it is
appropriate, the moral to be drawn is not necessarily one about the state
in question; it may equally well reect the poverty of our ordinary notion
of knowledge, insofar as it cannot subsume prima facie cognitive states of
this type, attitudes toward propositions or concepts.56
Objection: Is it your claim that metaphors are demonstratives or
indexicals, or that they are simply like them in some general respects? The
latter might be true, but it is uninteresting. The former obviously does not
t ordinary usage (unless `metaphors are demonstratives/indexicals' is a
metaphor!), and, despite the parallels you have pointed out, there are
great dierences between metaphors and demonstratives. It is simply im-
plausible to consider them one class of expressions (or interpretations)
falling under one theory.
On the one hand, demonstratives and indexicals have explicitly statable
rules that determine denite direct referents relative to their contexts of
utterance. Each utterance of `I' directly refers to its speaker or the agent
254 Chapter 6

of the context, an individual. Similarly, `now' and `here' each has a de-
nite value in each context. These rules are what we take to be the mean-
ings of indexicals and demonstratives. Hence anyone who masters the
language acquires knowledge of these rules, enabling her to assign the
expression a referent/content in each context without any additional spe-
cial skill or knowledge. Therefore, there is good reason to take such
knowledge of demonstratives and indexicals to fall under a semantic
theory.
On the other hand, is there anything analogous for metaphor? Even
your own candidate rule (Mthat) (see ch. 4, sec. I) is much less denite
and determinate than the rules for the indexicals. The rule that each token
of the metaphorical expression `Mthat[F]' expresses as its content prop-
erties P presupposed to be m-associated with F leaves unexplained what it
is to be m-associated with each F. It is also not straightforward to apply
the rule, even given an account of what it is to be m-associated. As you
indicated back in chapter 4, we must still select from among the m-
associated properties those that are appropriate in the context. There are
no rule-governed procedures to instruct us how to make this selection or
how to compute the nal value of the metaphorical character. Indeed,
selection for appropriate properties, and more generally the kind of com-
plex interpretation that metaphor demands, requires insight, guesswork,
intuition, tting or accommodating features to one another, skills of de-
tection and discovery not captured by rules and certainly not by rules of
language. Does anyone who masters a language plausibly acquire as part
of her semantic knowledge rules that suce to interpret a novel, context-
sensitive metaphor (even in its context)?
Reply: First of all, my claim is that metaphors and demonstratives
(including the full range of proper demonstratives and indexicals) fall
under one natural semantic kind; I am not sure whether that means that
metaphors are demonstratives or just like them. In either case, to say that
is not to deny that there are also signicant dierences, and especially
pragmatic dierences, between them.
But before I turn to those dierences, and their signicance, I want to
clarify (at the risk of repeating myself ) two potential confusions. The
objection claims that anyone who masters a language acquires knowledge
of the character-rules of indexicals and demonstratives that enable her to
assign them referents or contents in each context. In contrast, the skills of
interpretation (including intuition, insight, guesswork) involved in meta-
phor cannot be expressed by rules that could both assign metaphorical
interpretations and count as semantic, part of what every speaker of a
Metaphorical Character and Metaphorical Meaning 255

language knows. First, the characters both of demonstratives and index-

icals and of metaphors, like meanings generally, express linguistic con-
straints on possible contents of interpretation. Because they must be
completed by their respective contextual parameters, their (nonconstant)
characters or meanings do not themselves self-suciently generate any
contents or referents for their expressions; they merely exclude impossible
candidates, things that cannot be expressed by the words whose character
they are. Even someone who masters indexical language is not, then,
thereby enabled to assign referents or contents to the expressions apart
from context. The same is true of metaphors, but, by the same token, they
are no worse o with respect to their constraints. (To be sure, their
respective constraints dier substantively; the demonstratives and indexi-
cals refer to actual speakers, times, locations, and demonstrata, whereas
metaphors express properties presupposed to be associated with the literal
vehicle.)
Second, when a speaker knows the character of an indexical, she knows
a rule that constrains the referent/content of the indexical in a context;
but knowledge of that character-rule still does not furnish her with
knowledge of its referent/content unless she also knows who is speaking,
the time and location, and so on, knowledge of the context that is extra-
linguistic. Don't think that just because the content of a fully determined
indexical is determined by its character that the semantic knowledge of the
latter is also sucient for knowledge of the former. By the same token, it
cannot count against the determinate status of the character rule (Mthat)
for a metaphor that knowledge of it is not sucient to know its content,
for which we must also know its extralinguistic context, that is, context set
of presuppositions. Hence neither for demonstratives and indexicals nor
for metaphors is semantic knowledge of their respective characters su-
cient for speakers to have knowledge of their assigned contents.57
Turning back now to the dierences between the demonstratives and
metaphors, I want to argue (i) that the dierences between them are of
degree rather than of kind, and (ii) that there are also signicant dier-
ences within the class of proper demonstratives and indexicals. The dif-
ferences between demonstratives/indexicals and metaphors are easy to
overstate because we tend to focus on the most determinate indexicals at
one pole and the most indeterminate metaphors at the other. But when we
compare the full range of metaphors and demonstratives, we nd a con-
tinuum of context-sensitive interpretations. Pure demonstratives may be
no less indeterminate than some metaphors and the metaphorical inter-
pretations of many expressions are far more determinate than it rst
256 Chapter 6

appears. I'll argue for this by pushing from both directions: from below
from the alleged determinacy of demonstrativesand from abovefrom
the alleged indeterminacy of metaphors.
In discussing these examples, it is important to distinguish between how
determinate (as opposed to vague) is the content of the interpretation and
(keeping in mind that characters do not assign but constrain possible
interpretations) how the content is determined by the rule of character,
even though these often hang together.58 `I', for example, has a determi-
nate content (an individual, possibly at a time), and typically its content is
also fully determined by its characteralthough even for `I', there are
problematic applications, for example, its answering-machine uses in `I
am not here now'. When we move on to `here' and `now', as well as `you'
and `we', the relevant ``unit'' of content suers from vagueness or inde-
terminateness about the interval of its ``neighborhood'' (Is `now' the sec-
ond, minute, hour, day, or an even longer interval?), although the rule of
character is still fully determining. And when we move from the indexicals
to the complete (singular) demonstratives, not only may the indetermin-
ateness of the content grow (depending on what is demonstrated); the rule
of character, which requires a completing demonstration (either or both
presentation and ostensive gesture), is signicantly less determining. For
example, we inevitably point at an indenite number of things when we
point at any one thing. Hence it is arguable that all complete demonstra-
tives require a sortal modier, one that is often not verbalized, leaving
open a range of alternatives among which it may not be clear which (or
whether any) was specically intended by the agent. Add to that the nat-
ural imprecision of many ostensive gestures and the topiclessness of pre-
sentations. With dthat-descriptions, there can be a further indeterminacy
of character if the description is incomplete, failing to x a unique referent
in the actual context. Moreover, identication of the referent depends on
who/what is known (or believed) to satisfy the description, knowledge
that may be as allusive as knowledge of shared presuppositions. Finally, if
we turn from the singular to predicate demonstratives (e.g., `thus', where
it is a property that is ostended) there are all the old theoretical worries
associated with the individuation of properties in addition to the many
practical diculties in applying the rule of character. Here matters are not
much better o than with metaphor.
In short, there is a continuum of demonstrative and indexical expres-
sions, or interpretations of expressions, of more or less indeterminacy. At
one extreme there are the singular indexicals, at the other the predicate
demonstratives. Despite the fact that we clearly do have semantic knowl-
Metaphorical Character and Metaphorical Meaning 257

edge of their characters, in all these cases this knowledge needs to be

supplemented by, and to some degree cannot always be distinguished
from, the same kinds of insight, guesswork, and skill in rules of thumb
that we tend to associate with metaphor.
At the same time, we should not deny or even understate the fact that
metaphorical interpretation involves, as I tried to show in chapters 4 and
5, many non-rule-governed skills and abilitiesthe skills and abilities that
come into play, for example, in identifying and working out the networks
to which the vehicle of the metaphor belongs. This is precisely what a
pragmatics of metaphor must capture, although it is unlikely for the very
reasons cited by the objection that these skills and abilities can be put into
the form of rules or a theory (see ch. 5, sec. I). My aim is not to ignore
these non-rule-governed aspects of metaphorical interpretation but to ar-
gue that, because of them, we also should not lose sight of the semantic
constraints that are equally important. These constraints, captured by
characters that we know as part of our semantic competence, are what
become evident when, as with indexicals, we focus not on the interpreta-
tion of one token in a context but on the systematically related inter-
pretations of dierent tokens of one type in dierent contexts (see chs. 1,
4, and earlier in 6). When we ascend to this semantic vantage point, we
observe regularities and systematic constraints relating contents to their
respective contexts no dierent from those of indexicals. And as with
indexicals, the most plausible explanation is that our knowledge of these
constraints and rules constitutes our semantic competence in metaphor. In
these (and only these) respects I claim that metaphors and indexicals form
one natural semantic kind.
Objection: On your view that metaphors express propositional contents,
given that propositional contents are the very sort of things that admit
literal articulation (at least in principle), shouldn't it then be possible to
express literally everything a metaphor conveysin your terminology, all
knowledge by metaphor? And isn't this at best true of trite, common, or
dead metaphors, not novel, live ones?
Reply: No, it does not follow on my account that all knowledge by
metaphor is expressed in its propositional content, let alone that it can all
be stated literally. To explain why not, I now turn to chapter 7 and to our
knowledge by metaphorical character.
Chapter 7
Knowledge by Metaphorical
Character

I Marie's Problemand Ours

Marie, a young woman in her teens, suered from the eating disorder
anorexia nervosa. In treatment, she explained to her therapist that her
mother had forbade her to continue seeing her boyfriend. Angrily, she
reported, she had said to herself:
(1) I won't swallow that [referring to her mother's interdiction].
Let's assume that in the context in which she uttered (1) it was clear that
Marie's use of the word `swallow' was metaphorical.1 Let's also suppose
that what Marie said, the content of (1) interpreted metaphorically, is
expressed by
(2) Marie won't obey her mother's interdiction.
Does (2) adequately express everything said by Marie's utterance of (1)?
Does it exhaust the information conveyed by her utterance? Yes and no.
Yes, insofar as (1) is true, spoken by Marie referring to her mother's in-
terdiction, if and only if Marie does not obey her mother's interdiction.
No, insofar as her utterance of (1) is meant to contribute to an explana-
tion of her anorexic behavior, albeit as an irrational way of resisting her
mother's command. For to explain why Marie stopped eating in terms of
a belief we would ascribe to her on the evidence of her utterance of (1), we
must somehow include as part of the representation of her belief the fact
that what she said, namely, that she would not obey her mother, was
expressed metaphorically using the verb `swallow'. Only under that meta-
phorical mode of expression of what she saidonly if we include how
she metaphorically believed, or expressed, what she believedcan we see
any connection, conscious or unconscious, between her belief and her
subsequent anorexic behavior. To be sure, Marie's behavior and the
260 Chapter 7

connection she made are not rational, and no explanation should make it
so. But only by acknowledging the cognitive and explanatory signicance
of the metaphorical way in which she expressed her belief can we explain
her behavior at all.
This example may not be as innocent as we might like, but it gives us a
glimpse of how a metaphorical use or interpretation of language can
convey a kind of information, or bear cognitive signicance, above and
beyond what we might all agree is what it says, its propositional content
(in a context).2 The metaphorical mode in which Marie expressed her
belief is essential, not to determine whether what she said is true or false,
but for our folk-psychological purposes of explaining her behavior. In-
formation of this kind, conveyed specically by the metaphorical mode of
expression of a word interpreted metaphorically, is an example of what I
called, back in chapter 1, ``knowledge by metaphor.''
Here is a second, more innocent example of knowledge by metaphor,
taken from the Book of Samuel II, 12.3 After David has Uriah killed in
battle in order to cover up his indiscretion with Bathsheba whom he
then took as his wife, God sends Nathan the Prophet to reprimand him.
David, as far as we can tell from the text, feels little regret, shame, or guilt
for the act he committed. He is now married to Bathsheba, and he seems
to dismiss and forget the gravity of his oense. Nathan tells him the story
of a rich man and a poor man. The rich man has countless ocks of sheep
and herds of cattle, the poor man only one little lamb whom he loves as
dearly as his own child. When a traveler comes to town, the rich man
takes the poor man's one little lamb to feed the guest rather than use one
of his own large ock. Hearing this, David explodes with anger. He tells
Nathan that the rich man should be killed as punishment and that he
should compensate the poor man fourfold. At this point, Nathan points
his nger at David and tells him (here I translate with a little liberty):
``You are the rich man.'' David then confesses his sin and repents.
Now, what has David learned when Nathan tells him, using what
I take to be a metaphor, ``You are the rich man''? Surely Nathan has
not told David anything he did not know already (in at least one familiar
sense of ``know'')that he has all the wives and women that any man
could wish for, in contrast to Uriah who had only Bathsheba whom he
dearly loved; that it is wrong to take another's property, let alone his
beloved wife; that it is surely wrong to have someone else killed to save
one's own honor and satisfy one's own desires. What, then, is it about
this metaphoror the knowledge somehow carried by the metaphor in that
contextthat brings about this radical change of feeling and action in
Knowledge by Metaphorical Character 261

David?4 Whatever the specic information consists in, I want to argue

that it is knowledge by metaphor.
In chapter 1, I pointed out that most contemporary theories hold
that our knowledge of metaphor and knowledge by metaphor go hand in
hand. Those who hold that a speaker's knowledge of metaphorical inter-
pretation is part of his linguistic competence also hold that any meaning
or concept he is able to express by a metaphor should belong ``in princi-
ple'' to the stock of linguistic, or literal, meanings or concepts he knows in
virtue of his knowledge of language (and specically lexical knowledge).
Therefore, whatever can be expressed metaphorically ought to be equally
well expressible literally; no additional knowledge ought to be communi-
cated specically by the metaphorical mode of the metaphor that could
not be communicated by some literal utterance. On the other hand, those
who hold that someone's knowledge of metaphor is something other than,
or in addition to, his general linguistic competence at least allow for the
possibility that some kinds of information or knowledge conveyed by
some metaphors might not be expressible by a literal expression. And in-
deed those who hold that our knowledge of metaphor violates rules that
constitute our linguistic competence may even be committed to a stronger
position, according to which it is impossible to express literally any
knowledge by metaphor. Both of these last two positions acknowledge a
kind of knowledge conveyed specically and essentially by the meta-
phorical mode of expression of a metaphor.
My own view, which I promised back in chapter 1 to argue for later in
this book, is that the component of a speaker's ability to interpret a meta-
phor that falls within his semantic competence proper is precisely what
enables him to express knowledge by the metaphor that is not equivalently
expressible except through its metaphorical mode of expression. The time
has now come to pay o my promissory note. I shall argue in this chapter
that it is our semantic knowledge of the character of a metaphor that
enables us to express knowledge and information by the metaphor in ad-
dition to that expressed in its (propositional) content (in a context). The
propositional content of a metaphor, like any propositional content, can
always be expressed literally ``in principle,'' that is, given the availability
of a suciently rich literal vocabulary. But the ``character-istic'' knowl-
edge or information carried by the metaphorthe knowledge or infor-
mation that depends on and should be individuated by the character of
the metaphoris additional to the propositional content of the metaphor
(in a context) and is not always equivalently conveyed through other lin-
guistic means. Even if what the metaphorical character conveys can be put
262 Chapter 7

dierently, how it conveys that information is also signicantand not

literally duplicable.
Indeed, as we'll see toward the end of the chapter, even our formal
semantic notion of character is not always ne enough to capture all
the philosophically intuitive dierences of signicance we wish to register
between modes by which metaphors present their contents. To that extent
the ``character-istic'' information we wish to express outruns the expres-
sive power of our formal notion of character. But, as we saw earlier at the
end of chapter 3, section VII, similar shortcomings of the formal appara-
tus emerge in the theory of demonstratives. These reect as much the
limits of formalization as our theory in particular.
I'll begin by clarifying the problem posed for our knowledge by meta-
phor by a semantic theory of our knowledge of metaphor. In the past this
problem has usually been presented in terms of literal paraphrasability.
But this way of framing the issue has tended to obscure rather than clar-
ify. In the remaining sections of this chapter, I'll try to recast the problem
in a dierent mold.

II The Rise and Fall of Literal Paraphrasability

The theme of knowledge by metaphor is not new to the literature. It has

been taken up by way of a number of subjects: the cognitive signicance
(or insignicance) of a metaphor; the surprise occasioned by a metaphor;
and its gurative, or pictorial, versus its descriptive aspects. However, by
far the most attention has focused on whether metaphors are literally
paraphrasable: whether the meaning of a metaphor can be equivalently
stated in literal words. This emphasis is not surprising in light of the his-
tory of theories of metaphor, but it has not been helpful in furthering our
understanding of the nature and scope of the problem.
Historically, it appears to have been the rhetorical origins of the inves-
tigation of metaphor that gave rise to the idea that metaphors are para-
phrasable. Beginning with Aristotle (though this was hardly his full view)
through the rhetorical tradition in antiquity and the middle ages, a meta-
phor was mainly viewed as a decoration or ornamenta matter solely or
primarily of formsuperadded to the content of language, which, it was
thought, could always be extracted from the metaphor and then expressed
literally. Thus the Andalusian Hebrew poet and literary theorist Moses
ibn Ezra (c. 1055 to after 1135) wrote in praise of biblical metaphors:
Know that . . . metaphor is the most beautiful [of the literary gures] for both
[poetry and prose]. And although that which is stated explicitly is fundamentally
Knowledge by Metaphorical Character 263

more accurate whereas metaphor is an irregularity, [the latter] is nevertheless

possessed of elegance. When discourse is covered with the garments of metaphor
and embellished with gure, its silken embroidery is made comely and its enam-
eling, beautiful. The dierence between ornamented and naked discourse is like
the dierence between eloquence and stammer. . . . Whoever among the contem-
porary men of understanding rejects this [use of ] metaphor, denigrates the certain
and the obvious and departs from the straight path. . . . For metaphor is used fre-
quently in our [sacred] texts. . . . There is no harm in this; indeed it is inescapable.5
Here a metaphor is said to provide beautiful clothing for the naked body
of discourse, its ``unembellished'' content, which we can identify from its
expression in ``naked discourse'' or plain speech, bare of all ornaments
and decorations. As the modern critic A. E. Housman says, metaphor and
simile ``are inessential [even] to poetry''; they are ``accessories,'' employed
by the poet ``for ornament'' only because the image contained possesses
an ``independent power to please.''6 If we strip o the metaphorical sur-
face of an utterance or inscription, there exists beneath it a statement that
would express its content explicitly. It is this possibility of an explicit lit-
eral paraphrase that underwrites the distinction between form and content
presupposed by the rhetorical conception of metaphor as a purely stylistic
device.
In Renaissance rhetoric, the ornamental conception of metaphor seems
to have meant something dierent. Brian Vickers explains that while
gures were described as ``ornaments of rhetoric,'' an ornament was not,
as we mean nowadays, ``a decoration not functional to the overall aim,''
but rather a piece of equipment or accoutrement, an instrument to make
discourse more intense and eective, even if primarily emotionally. Thus
gures were ``functional, persuasive, not decorative.''7 It is, perhaps,
against this background that we should also understand the famous neg-
ative comments of Hobbes and Locke, who charge that the essentially
rhetorical function of metaphor makes it dangerous, deceptive, and un-
desirable, something to be eliminated wherever and whenever possible.8
The rhetorical conception of metaphor as ornament, in any of these
versions, is rarely held today. However, its noncognitivist motivations
survive in the guise of two contemporary cognitivist arguments for literal
paraphrasability. The rst of these is semantic eliminativism. If we can
state everything the metaphor says in a literal paraphrasesay, by sub-
stituting literal for metaphorical terms or by expanding the metaphor into
a literal comparison statementwe can eliminate the need to posit a
special kind of metaphorical meaning. Instead we can explain how meta-
phor works either as a species of our knowledge of the literal or as one
kind of speaker's meaning. The second argument originated as a reaction
264 Chapter 7

to the view that metaphors are merely expressions of feeling or emotion

that lack all cognitive content, a view that was held by movements as
disparate as the Romantics and logical positivists. In response to these
positions, in order to legitimate metaphor as the expression of ``serious
thought'' (Black 1962, 25), it was argued that they make truth-claims no
dierent than literal language. But where there is truth and falsity, there
must be truth-conditions or propositional content. Assuming that the lan-
guage contains a suciently rich literal vocabulary, any such metaphori-
cally expressed propositional content should therefore be ``in principle''
literally expressible. For, despite its metaphorical means of expression,
what the metaphor expresses should be the same kind of language-
independent or language-neutral propositional content that can be ex-
pressed literally. After all, the very nature of propositional content is
to be an abstracted object of understanding that admits, and can be
equally well expressed by, dierent inter- and intralinguistic expressions.9
All metaphors should therefore be literally paraphrasable.
The intentions of these proponents of metaphorical cognitivism were
honorable, but the net eect of their arguments was to rearm the old
view that there is no increment of information added to the literally ex-
pressible propositional content of a metaphor by its metaphorical means
of expression. The metaphorical mode of expression turned out, once
again, to be of no cognitive value, even if the function of metaphor was
no longer thought to be essentially rhetorical, ornamental, or decorative.
It was in reaction to such ``decoration'' theories of metaphor that gures
like I. A. Richards, Cleanth Brooks, and Max Black, in the 1950s, insisted
on the cognitive autonomy of at least some metaphors.10 As Black put it:
Suppose we try to state [their] cognitive content in ``plain language.'' Up to a
point we may succeed. . . . But the set of literal statements so obtained will not
have the same power to inform and enlighten as the original. . . . One of the points
I most wish to stress is that the loss in such cases is a loss in cognitive content. . . .
[The literal paraphrase] fails to be a translation because it fails to give the insight
that the metaphor did. (1962, 46)

Black defended his view by proposing on the same page his own ``inter-
action'' theory of metaphor. However, even if we reject his defense, there
is no denying that he identied a compelling intuition that (at least some)
metaphors are, for some reason, unparaphrasable. No matter how com-
plete, how detailed, how subtle a literal paraphrase one provides for a
particular metaphorindeed the more detailed and fuller, the longer, the
more prolix the paraphrasesomething seems to be invariably lost, at
least a signicant dierence in structure and eect, however hard it is to
Knowledge by Metaphorical Character 265

spell out the relevant idea of structure and in what this eect consists.
Hence no literal expression can serve as a ``full'' paraphrase of the cogni-
tive content of a metaphor and thereby substitute for it.
I share Black's forceful intuition, but it remains for us to make sense of
this dierential information. The cognitive autonomy theorist lumps all
dierences between the metaphorical and literal into one category he
labels cognitive content. But when asked to be more precise and explicit
about the dierence in cognitive content between the metaphorical and
the literal, he is pushed to more and more obscure defenses. Black, for
example, says that the ``insight'' fostered by the metaphor that is not
expressed by its literal paraphrase results from ``a distinctive intellectual
operation'' (ibid.). However he immediatelyand paradoxicallyadds
parenthetically that this operation is ``familiar enough through our
experiences of learning anything whatever'' (ibid.)! One quickly begins
to suspect that the autonomy theorist's view of the distinctive insight of a
metaphor is less and less distinguishable from the view that what meta-
phors distinctively express is really nonpropositionalfeeling, emotion,
or (in Frege's terminology) coloring. In sum: Neither of the two tradi-
tional campsthe decoration theory or the cognitive autonomy theory
do justice to both intuitions and theory. We end up with either descrip-
tively inadequate paraphrases of metaphors or hopelessly obscure expla-
nations of their unparaphrasability.
Consequently, the general dilemma for accounts of knowledge by
metaphor is to show how the following two theses jointly hold.
(A) Metaphors (i.e., utterances in which at least one constituent is
interpreted metaphorically) are truth-valued utterances; hence metaphors
have truth-conditions or propositional content no dierent from that of
literally interpreted utterances.
(B) The information or knowledge or cognitive content (i.e., content
that is either true or false, or true or false of things) communicated by a
metaphor is at least in part a function of the specically metaphorical
mode by which the utterance is interpreted.
Contemporary theorists have reacted to this dilemma in dierent ways.
Davidson's reply, for example, is to defend (A) by denying (B). As we
have seen, he argues that a metaphorical utterance expresses nothing
other than what the utterance expresses with its literal meaning (according
to which it is typically false). The purported information cited in (B) is
really an eect of the utterance rather than content ``contained'' in it, and
typically it is not even propositional. A second extreme, taken (in dier-
266 Chapter 7

ent ways) by Lako and Johnson (1980) and by Ricoeur (1962, 1978),
appears at rst to defend both (A) and (B). However, they also claim
that the received notions of truth and propositional content assumed by
(A) are based on an illegitimate paradigm of literal meaning and objec-
tive truth and must, therefore, be rejected. In eect they defend (B) by
denying (A).
My own position assumes (A), the thesis that (assertions of declarative
sentences containing) metaphors are truein the same sense of `true' in
which `snow is white' is true. And some of the arguments in this chapter
will lend this assumption additional indirect support. But my main aim is
to show how my semantic conception of our knowledge of metaphora
conception built on (A)supports (B): how the very same semantic com-
petence in metaphor underlies the speaker's ability to express knowledge,
or information, by the metaphor that is not expressed as part of a literal
statement of its propositional content in its context of utterance.
Before turning to this positive project, one last preliminary comment: I
alluded several paragraphs back to reservations about the usefulness of
literal paraphrasability as a device to explain our knowledge by meta-
phor. Let me say a word about my doubts. To begin with, the very idea
that literal paraphrasability should be a necessary condition for a meta-
phor to have cognitive content or to convey information undercuts the
function of specic metaphors. As daily experience demonstrates and
as historians of language remind us, metaphors are frequently used to
express contents for which no literal expression is available at the time
of utterance. So why require that all metaphors always be literally para-
phrasable?11 Furthermore, it is never made clear what constitutes an
adequate literal paraphrase of a metaphor. Must the paraphrase be a
single simple expression or can it be a complex string or phrase? Apart
from the fact that which concepts are lexically represented by simple
expressions and which by complex phrases is itself an idiosyncratic fact
that varies from language to language, the rst alternative would seem
too strong, the second too weak.12 Finally, there is a deeper problem
with literal paraphrasability as a litmus test for the cognitive status of a
metaphor. The condition achieved its prominence against the background
of a set of assumptions about meaning and linguistic competence that
prevailed earlier in the century among Anglo-American philosophers of
language. It was assumed that each competent speaker has complete,
context-invariant, and determinate understanding of all the meanings of
all the expressions in his language; that this knowledge of meaning con-
sists in knowledge of truth- (or satisfaction-) conditions; and that a com-
petent speaker should be able either to articulate the cognitive meaning of
Knowledge by Metaphorical Character 267

any expression in an explicit denition or synonymous paraphrase or,

where that is not feasible, to manifest his knowledge of meaning by dem-
onstrating that he recognizes when sentences (containing the expression)
are established on particular occasions to be true or false.
Few philosophers of language nowadays would accept all of these
assumptions, and many would accept none of them. The requirement that
the cognitive meanings of expressions should be generally denable and
paraphrasable is subject to many of the same kinds of criticisms leveled
by Kripke and others against description theories of proper and general
names. Quine's critique of synonymy and analyticity back in the 1950s
undid the explanatory usefulness of talk of paraphrase as part of accounts
of knowledge of meaning. The identication of meaning with truth-
conditions does not hold in general, as we have seen with demonstratives.
And, last of all, the assumption that a speaker's linguistic competence
consists in complete and determinate knowledge of meanings rests on an
unacceptably strong individualistic conception of mind and knowledge. I
shall not review objections to this view, but my account gives more cen-
tral place to the extralinguistic context in individuating the contents of
one's knowledge by metaphor.13 In short, the ability to articulate explic-
itly in a (literal) paraphrase what one knows when one understands a use
of language is not to be expected even for literal language. On pain of
enforcing a double standard, we should not demand more of metaphor.
These reservations do not show, of course, that literal paraphrasability
has no value in understanding knowledge by metaphor. Rather my point
is to displace it from center stage, to situate it within a broader arena of
discussion that includes other ways in which knowledge by metaphor is
manifest. The rst claim I shall discuss, the ``endlessness'' of metaphorical
interpretation, was traditionally oered in support of the unparaphras-
ability thesis; in recent years philosophers have employed it to draw a
diametrically opposed conclusion about the nature of metaphorical
meaning. However, neither side of the dispute has paid sucient attention
to the nature of the purported endlessness. To get a better look at this
feature, we must also look at other ways in which knowledge by metaphor
is displayed. In the next section I'll take the rst step toward these other
approaches to the topic.

III The Endlessness of Metaphorical Interpretation

According to a widely shared intuition, metaphors admit endless or at

least open-ended interpretations; what they mean or communicate is
never antecedently xed. William Empson describes metaphors as ``preg-
268 Chapter 7

nant'' with meaning. Stanley Cavell calls a metaphor ``burgeoning'' with

meaning; he insists that paraphrases of metaphors signicantly end with
`and so on'.14 Davidson writes that what a metaphor conveys is not
nite in scope. . . . [T]here is no limit to what a metaphor calls to our attention. . . .
When we try to say what a metaphor `means', we soon realize there is no end to
what we want to mention. If someone draws his nger along a coastline on a map,
how many things are drawn to your attention? You might list a great many, but
you could not nish since the idea of nishing would have no clear application.
(WMM, 263)
However, despite their common premise, these authors draw opposing
conclusions. Empson and Cavell conclude that metaphors must have a
special kind of cognitive content or meaning. If metaphors are true,
they must be ``wildly truemythically or magically or primitively true''
(Cavell 1967, 81). Davidson, on the other hand, infers from the fact that
they lack a nitely speciable set of truth-conditions that metaphors can-
not be used to make true or false assertions in the received senses of
``true'' and ``false''; hence they have no cognitive or propositional content
(other than their literal one).
Nonetheless all these authors seem to agree that in fact there is no end
to the number of properties, features, or things a metaphor expresses. But
what is the nature of this purported ``innitude,'' ``endlessness,'' or ``in-
exhaustibility'' of whatever it is that a metaphor conveys? Before we draw
conclusions from a phenomenon, we must secure its existence.15
One relatively noncontroversial way in which a metaphor (type) F
admits limitless interpretation is that there is no xed upper limit on the
number of dierent contents (tokens of ) F expresses in dierent con-
textsrelative to dierent context sets of presuppositionsbecause there
is no upper bound on the number of dierent contexts in which F can be
used. Recall, again, our examples (1)(10) in chapter 1, section II. In each
example, one character for a single metaphorical expression (type), say,
`Mthat[`is the sun']', determines dierent contents in the dierent con-
textsgiven dierent sets of presuppositionsin which it is uttered. Thus
a single metaphor (type) is capable of apparently endless or at least an
indenitely large, unbounded number of interpretations, contents, in its
respective contexts. However, apart from knotty questions about how to
individuate contexts, note that in each respective context, the interpreta-
tion, or content, of the metaphor is nite and xed. The conditions
under which the utterance is true or false in that context are determinate
and nite. If we know the context set and our literal vocabulary is rich
enough, it even ought to be possible for us to state the content nitisti-
Knowledge by Metaphorical Character 269

cally. Hence this kind of limitlessness of metaphorical interpretation nei-

ther conicts with the nitistic requirements of a semantics (as Davidson
seems to argue) nor requires a dierent kind of truth or content sui generis
to metaphor (as Cavell suggests).16
This explanation also suggests an additional way in which the knowl-
edge conveyed by a metaphor is not exhausted by its xed, nite content
in a single context. Besides knowing the specic content each utterance of
the metaphor generates in each context, the speaker also knows (in each
context) that its context-sensitive character would determine a dierent
content given a dierent context. And he knows that if he knew what the
alternative presuppositions were, he could also gure out, or calculate,
what its content would be. This knowledge, corresponding to the charac-
ter of the metaphorical expression, resembles the ``directions'' contained
in the characters of indexicals like `I'.17 Both instances of character-istic
information endow their respective types with a power or potential for
further, future interpretation lacked by the characters of expressions
(types) that are context-independent. But for both types of expressions,
indexicals and metaphors, the ``directions''how to interpret them in
arbitrary contextsthat we know in knowing their characters are never
themselves part of their contents in any particular contexts; therefore they
will not be expressed by any (literal) statement of the content of an
utterance, be it a metaphor or indexical, in a particular context. This, I
propose, is Cavell's insight when he insists on the signicance of the ``and
so on'' that comes at the end of any example of a literal paraphrase of a
metaphor. Obviously this ``and so on'' cannot be eliminated by a specic
supplementation of content, no matter how much content we add. What
the ``and so on'' signals is a potential for a dierent or additional inter-
pretation that, by hypothesis, no actual interpretation can capture or
exhaust.
This potential for further interpretation is essential in order to under-
stand properly how metaphors function in science. One scientist proposes
a metaphor based on a content it expresses in an initial context, a context
in which, say, he notices, and so (in our sense) presupposes, that the literal
referent of the vehicle exemplies some feature to be attributed. What he
says, relative to that presupposition set, is taken to be true or false in that
context. But in proposing the metaphor, he may also intend for others to
adopt it as their own: to accept the truth expressed by the metaphor (in
that context) as a hypothesis in order to explore it, both by extending the
metaphor in subsequent contexts relative to expanding presupposition
sets built on his own and by following out the associations made by the
270 Chapter 7

various networks to which the literal vehicle belongs.18 However, notice

that the content of the metaphor in future contexts is neither determined
nor constrained by the original author's intention in choosing the meta-
phor with its respective character, or meaning. In the course of exploring
the metaphor in subsequent contexts, later interpreters are not revealing,
or recovering, content originally intended by the one who introduced the
metaphor. Once proposed, the metaphor possesses a life of its own, its life
span limited only by the degree of potential carried by its character and
by the presuppositions that depend on the creativity, imagination, and
powers of discovery of its users in their respective contexts. Metaphors
may be pregnant with meaning, but that need not mean that all po-
tential interpretations are ``embryonically'' contained in the initial content
or in the original Sinn in the mind of the one who rst conceived the
metaphor.
Yet these explanations for the endlessness of metaphorical interpreta-
tion are not the end of the story. What many writers, Davidson and
Cavell included, have in mind is not only the countless number of dier-
ent interpretations that a single metaphor (type) might have in dierent
contexts. Nor is it the potentiality for further interpretation carried by the
character of the metaphor. They also want to claim that what a metaphor
expresses within one context has no end. This fact, if it is a fact, does
conict with the requirements of a nitistic semantic theory, thereby con-
stituting a major dierence between metaphorical and other kinds of
semantic interpretation. And, as we saw earlier, dierent authors draw
dierent morals from this alleged intracontextual interpretive endlessness.
For some, so much the worse for nitistic semantics. For others, so much
the worse for metaphor, or for metaphorical meaning, or for the possi-
bility of a semantics of metaphor.
But what could this kind of intracontextual interpretive endlessness
consist in? Are these authors claiming that if we actually try to state the
content of a particular metaphor, we in fact nd ourselves going on and
on, mentioning feature after feature, proposition after proposition, liter-
ally never coming to an end? As a matter of practice, I know of no evi-
dence (fortunately) that this is ever true.19 To borrow the metaphor of a
few paragraphs back: metaphors may indeed be pregnant with meaning,
but it would also be a strange pregnancy if it had no term and if it issued
forth in endless streams of progeny.
Although we have not yet found evidence that metaphorical inter-
pretation is endless in an objectionable way, there is nonetheless a
widespread sense that metaphor is semantically anomalous because (as
Knowledge by Metaphorical Character 271

Davidson put it) it is so hard so much of the time to say ``exactly''

determinatelywhat the content of a live, vital metaphor is supposed to
be, especially in poetry and literature. Can we explain (away) this appar-
ent endlessness or indeterminacy in some other way?
One account would begin with the explanation of how context or,
more precisely, the force of the utterance xes determinate content for
other uses of metaphor. An utterance that purports to make an assertion,
or to communicate information (or to perform another speech act that
presupposes assertion), purports to have a determinate content or truth-
conditions. So, when a metaphor occurs in a prima facie assertion (or
utterance presupposing assertion), that use typically exerts pressure on the
participants to x their presuppositions in order to generate a determinate
content for the metaphor that must be satised for the utterance to be
true, to be assertible.20 But by the same token, if the utterance does not
itself have the force of or presuppose assertion, indeterminacy can be the
result. Metaphors in poetry or literature whose point is not assertion may,
on the contrary, aim at a multiplicity of alternative contents (and thereby
ostensible truth-values).21 Rather than pressure us to x and delimit
the presuppositions, these contexts may nurture the very plurality of pos-
sible contents that results when we leave a unique selection of the pre-
suppositions unresolved, a range of alternative sets of presuppositions left
open. For these metaphors, there may indeed be no one unique proposi-
tion asserted. But the coherence of such indeterminacy presumes that in
some context and used in another way, the content of a metaphor is xed
and determinate: that the content of the metaphor in some context does
uniquely determine one or another truth-value. The most plausible way,
in other words, to account for indeterminacy of metaphorical content
the diculty we sometimes have (and some critics and readers welcome)
in deciding on the content of some metaphorsis by assuming, at least
under a reasonable idealization, that there are metaphors with determi-
nate contents.
In any case, we have not yet discovered a sense in which our knowledge
by metaphor within a single context is ``endless'' in a way that would
conict with the requirements of semantic knowledge of metaphor. But
this is not the end of our story. In the following sections I shall turn to
other ways in which our knowledge by metaphor is manifest, including
other ways in which a literal statement may not capture ``everything''
conveyed by a metaphor. This will eventually lead us back to a dierent
dimension of the character of a metaphor in terms of which we will be
able to explain its ``endlessness.''
272 Chapter 7

IV Metaphorical Mode of Presentation

Throughout this book (e.g., in ch. 1, examples (1)(10)) we have seen how
one metaphorical character can determine dierent contents in dierent
contexts, that is, relative to dierent sets of presuppositions. Inversely,
one content (say, one property) can be determined by dierent meta-
phorical characters relative to dierent sets of contextual presuppositions.
For example, in Nigeria people use the metaphor ``She is my bedbug''
as a term of aection for their lover (or beloved)ostensibly because
bedbugs are cute there.22 Of course, given our presuppositions about
bedbugs, if we Americans were to ``translate'' the same metaphor, we
would be saying that she is a nuisance. To express the same content as
the Nigerians with a metaphor, we (or at least W. C. Fields) might say:
``She is my little chickadee.''23 In short: dierent contexts, same content,
dierent metaphorical characters.
This relation between metaphorical character and content is reminis-
cent of the relation between Frege's notions of sense and reference (see
ch. 3, secs. I, VI). Frege, you'll recall, posits the notion of sense in order
to solve his puzzle of identity (among other reasons): to explain how true
identity statements like
(3) The Morning Star The Morning Star
(4) The Morning Star The Evening Star
might dier in their cognitive signicance or informational value. (3) is
known a priori to be true; learning (4) might be and, for some ancient
Babylonian, presumably was a genuine empirical discovery. Frege's ex-
planation of this cognitive dierence is to distinguish between the refer-
ents of the terms and the dierent modes under which they present their
referents. The mode of presentation of the referent, which Frege locates in
the sense, or Sinn, of the expression, constitutes the qualitative perspective
from which the speaker is epistemically related to the thing. This epis-
temic dierence between the senses of the terms anking the identity sign
in (4) accounts for its informativeness, unlike the uninformative (3) whose
anking terms have the same sense as well as referent.
Frege's own examples involve proper names and denite descriptions,
that is, eternal singular terms. However, as we also saw in chapter 3, sec-
tion VI, Kaplan shows that the same puzzle arises with demonstratives:
(5) That [the speaker points at Venus in the morning sky] That [the
speaker points at Venus in the morning sky]
Knowledge by Metaphorical Character 273

(6) That [the speaker points at Venus in the morning sky] That [the
speaker points at Venus in the evening sky]
(where (6) is uttered very slowly). Because Frege's solution in terms of
sense won't work for these directly referential terms, Kaplan proposes
that we look to the dierent characters of the respective complete demon-
stratives, where those characters are individuated by their dierent dem-
onstrations or presentations. One presents Venus as seen in the morning
sky, the other as seen in the evening. Just as Frege views sense as the mode
of presentation of its referent, so Kaplan proposes that we view the char-
acter of a complete demonstrative as a mode of presentation of its content.
To pursue our explanation of metaphor on the model of demonstra-
tives, I want to propose that metaphorical character also provides a mode
of presentation of its content (in a context). That is, there is information
or cognitive signicance carried and individuated by the character of the
expression interpreted metaphorically, or its corresponding metaphorical
expression, above and beyond its propositional, truth-conditional content
in a context, ``character-istic'' information that is never captured in a
statement of the content alone. But from that it does not follow, let me
immediately add, that the character-istic information is dierent in cog-
nitive kind from the information contained in propositional content.
Everything else being equal (which, we shall see in a minute, is also not
entirely the case), there is no reason that information or cognitive value
should not be expressible by some (other) proposition; it is just not part of
the content generated by that utterance in its context.
If we turn back for a moment to the old problem of literal para-
phrasability, it is now tempting to try to recast it in the mold of Frege's
puzzle. Recall that the problem of paraphrasability was that, on the one
hand, if metaphors have propositional content, it ought (``in principle,''
given a rich enough literal vocabulary, etc.) to be possible to state that
content without change (loss or, for that matter, gain) in informativeness
in literal language. On the other hand, the information or knowledge
expressed by a metaphor also appears to be a function of its metaphorical
mode of expression, which is not preserved in literal paraphrases (as the
many descriptively inadequate literal paraphrases attest). Now, in light of
our parallel with demonstratives, why not treat that dierence in infor-
mativeness between a metaphor and its literal propositional paraphrase
on the model of our account of the dierence in informativeness between
the identity statements in the demonstrative version of Frege's puzzle?
Why not treat the dierence in information or cognitive signicance
274 Chapter 7

between a statement containing an expression interpreted metaphorically

such as
(7) Phil is an eel
or, equivalently,
(7*) Phil Mthat[`is an eel']
uttered in a context c and a literal statement with (let's suppose by hy-
pothesis) the same content in c, for example,
(8) Phil is stealthy,
just as we would treat the dierence between the two identities
(9) The proposition that Phil is an eel The proposition that Phil is
stealthy
or, equivalently,
(9*) The proposition that Phil Mthat[`is an eel'] The proposition that
Phil is stealthy
and
(10) The proposition that Phil is stealthy The proposition that Phil is
stealthy.
(where the context c in which (7)(10) are all uttered is a context in
which `is an eel' is interpreted metaphorically to express the same con-
tent expressed by `is stealthy' interpreted literally)? The same solution pro-
posed to account for the dierence in informativeness between the two
demonstrative identity statements (5)(6) with the same content should
carry over to the dierence in informativeness, or cognitive signicance,
between the statement containing a metaphor and the literal statement
with, by hypothesis, the same content. In both cases, the dierence in
informativeness is a function of their dierent characters, which present
their common content under dierent modes. It is precisely this character-
istic dierence that is felt to be lost in paraphrase.

V Surprise

Surprise is another feature associated with metaphor on which we might

also be able to throw some light using the Fregean account of informa-
tiveness in terms of dierences in mode of presentation. According to
Frege (1966), the relevant notion of informativeness that distinguishes (6)
Knowledge by Metaphorical Character 275

from (5) is manifest in the fact that it is a discovery for someone to learn
that identities like (6) are true, that such statements ``contain very valu-
able extensions of our knowledge and cannot always be established a
priori'' (56). Something similar happens when we hear or learn a meta-
phor (or a metaphor/literal identity like (9)). As philosophers since Aris-
totle have observed, some metaphors have a particular power to occasion
surprise.
Liveliness is specially conveyed by metaphor, and by the further power of sur-
prising the hearer; because the hearer expected something dierent, his acquisition
of the new idea impresses him all the more. His mind seems to say, `Yes, to be
sure, I never thought of that.' (Rhetoric 1412a 1821)

But not all notions of surprise are relevant for metaphor. One idea
makes it a function of the probability of the sentence (type) being tokened;
the lower the probability, the greater the surprise. This notion won't work
for us since there is no plausible way to assign probabilities to the tokening
of sentences on occasions.24 A second notion makes surprise a matter of
unpredictability. But we have argued that, constrained as they are, meta-
phors are not absolutely unpredictable. A more promising idea for our
purposes takes surprise to be a ``cognitive emotion'': an emotion that
presupposes that its subject has certain accompanying beliefs and expec-
tations that purport to justify it and that would be unjustied if the beliefs
and expectations turn out to be false.25 In the passage quoted, Aristotle
proposes two cognitive conditions for the surprise occasioned by a meta-
phor: (1) the subject must believe that he has acquired a ``new idea''
through the metaphor; and (2) his acquisition of the new idea must
somehow dier from his prior expectations. But what is the new ``idea'' he
must believe he has acquired with the metaphor, and how must it dier
from his prior expectations? When the hearer expresses his surprise by
saying ``I never thought of that,'' what does the demonstrative refer to? Is
it the content (in that context) of the metaphor: that Juliet has some
property Pthat she is worthy of worship and of Romeo's undivided at-
tention? No, what is new cannot simply be the unprecedented attribution
of the property in question to the subject; it cannot be a function simply
of the content of the utterance. If it were, the surprise occasioned by the
metaphor would be no dierent from what follows any novel application
of a literal predicatethat is, any literal attribution of a previously
unattributed property. This kind of surprise or novelty is not insigni-
cant but it is hardly specic to metaphor. Instead, the surprise must be
a function, at least in part, of its metaphorical mode of expression or
276 Chapter 7

attribution, that is, the character of the metaphor. What the hearer never
thought of is that Juliet is the sunnot literally, of course, for that is not
surprising. We all know that the (literally expressed) proposition is false.
But what may be surprisingboth new and dierent from a prior expec-
tationis that Juliet Mthat[`is the sun'], that she can be truly ascribed a
certain property under the mode of presentation conveyed by `Mthat[`is
the sun']' given the contextual presuppositions associated with the meta-
phorical expression. The property itself may be one we could express lit-
erally or it may be expressible only metaphorically. But even if we could
express the same content some other way, it would still be a substantial
cognitive accomplishment to see that we can express or refer to it by
employing that metaphor in that context. In either case, the new idea is a
function in part of the contribution of the character of the metaphor, the
context-specic perspective from which it enables us to grasp and express
the feature ascribed.
It is more dicult to say exactly what the hearer must have expected
dierently that contributes to its surprise when he hears the metaphor.
Once again, if the surprise in question is occasioned by the utterance be-
cause it contains a metaphor, it cannot be a dierence only between its
content (in the context) and prior beliefs. But the divergence from prior
expectations might also not be of one kind for all metaphors. One kind
of divergence would presumably be where the utterance expresses a
``semantically anomalous'' proposition, or category mistake, under what
would have been its literal interpretation. In that case, the surprise or
novelty would be, as Goodman (1976) puts it, ``a matter of teaching an
old word new tricks'' (69). In previous accounts, the point in appealing to
such literal anomalousness was to explain why the utterance is identied
as a metaphor; here I am suggesting that the anomalousness registers the
``distance'' (however that is measured) between the characters of the
expressions employed in the utterance, not to show how literally unlike
they are from each other, but to signal how unlikely it is that we would
antecedently think that we could say something true about the content of
the one using the other. As Aristotle says (ibid., 1012), a resemblance
perceived between ``things far apart'' is more striking, and hence surpris-
ing, than one between things closely related. On this view, the expecta-
tions need not contradict or conict with the metaphor; it is sucient
if the resemblance is simply unanticipated given prior expectations.26 In
either case, the departure from prior expectations is not what makes the
utterance a metaphor, but what makes the metaphorical interpretation
surprising.27
Knowledge by Metaphorical Character 277

VI Metaphorical Perspective

Can we now say more specically what the cognitively signicant dier-
ence between the characters of metaphorical and literal expressions con-
sists in? And how the metaphorical mode of presentation of a content (in
a context) is dierent from the content itself ? Let's begin with another
example from Romeo and Juliet. When Lady Capulet entreats Juliet to
take a greater interest in Paris, she tells her:
Read o'er the volume of young Paris' face
And nd delight writ there with beauty's pen.
Examine every married lineament
And see how one another lends content;
And what obscur'd in this fair volume lies,
Find written in the margent of his eyes.
This precious book of love, this unbound lover,
To beautify him only lacks a cover.
The sh lives in the sea; and 'tis much pride
For fair without the fair within to hide
That book in many's eyes doth share the glory
That in gold clasps locks in the golden story.
(I,iii, 8192; my italics)
Here Lady Capulet presents Paris to Juliet as if he were a book, drawing
on the metaphor network that people are books, a metaphor scheme that
was especially popular among the Elizabethans (and is still used today).28
People can be `read' like books, hence truly known by those and only
those who have `literate' skills. Their facial features are `signs' that `com-
municate' `content'. Eyes are `margins' in which `commentary' and `em-
phasis' are `written', clues how to interpret what is `hidden' in the `pages'
of the person's life. Its `covers' complete the book here; hence Paris with-
out a lover is `unbound', incompleteand, with some implied sexual
connotation, naked.29 Furthermore, as important as these many book-
like properties ascribed to Paris is the opening directive of the passage to
Juliet to `read' Paris. Not only are (some) people books; others are (re-
lated to them as) readers and authors. (It is not obvious what the sortal
alternatives to the `people are books [to be read]' metaphor would have
been for Shakespeare [`people are pictures to be viewed'?] but clearly the
scheme would be dierent, as it might be nowadays, if the alternatives to
`books' were `software' or `a computer program/le' or `a video'.)
These are the terms in which Lady Capulet metaphorically describes
Paris to Juliet. Were we to try to state the content of her speech (as liter-
ally as we can), it would be something like: `Study Paris closely, both his
278 Chapter 7

natural features and his behavior, and you'll discover his many natural
virtues, attractive qualities, his capacity to love. To learn his inner virtues
and character traits, pay close attention to his outward acts and fea-
turesand, realizing all this, you'll realize that all this wonderful man
lacks is a wife. Indeed the woman who marries him will share in his uni-
versal admiration and glory. Therefore . . .'.
Shakespeare does not express this content literally. Nor does he use a
set of independent individual metaphors. Instead he uses a single sys-
tematically intraconnected metaphor network. In chapter 5, I argued at
length that metaphor networks are essential to determining the contents of
the individual metaphorical expressions that belong to them. Here I want
to focus on a second role of metaphor networks that turns on the infor-
mation carried by the sums of their constituent characters. By employing
this network that determines the contents of its member expressions,
Shakespeare also adds a cognitive signicance to Lady Capulet's speech
in its entirety. But this additional signicance is not more content; rather it
is a perspective on, or a way of seeing, that content. Let me try to explain
what I mean by (the metaphors) a ``perspective on'' or a ``way of seeing or
presenting'' content, beginning with one class of nonmetaphorical expres-
sions: not surprisingly, indexicals.
Consider the context-oriented perspective that is a feature of indexical
language. A crucial part of what a speaker knows when she knows the
meaning, or character, of one indexical is knowledge of its interindexical
relations and the mandatory conditions under which one indexical must
replace another. For example, a speaker knows that when she utters a
sentence S of the form `. . . today . . .' on day d to express one proposi-
tional content, then in order to express that same content on day (d 1),
she cannot use S but must shift (assuming she wants to use an indexical)
to S 0 , which contains `yesterday' in place of `today'. If she does not know
that required transformation, she fails to know something essential to the
meaning of `today'. Thus one knows the meaning of any one indexical
only if one knows the meaning of the indexical system to which it belongs.
To borrow an image from James Higginbotham (forthcoming), the set of
indexicals constitutes a set of coordinate points marking their respective
parametric positions as they are interrelated in a context. When we
change perspective by referring to the same individual from a dierent
positioncalling today `yesterday' tomorrowto preserve the reference
we must shift, not just the one indexical, but its whole coordinate system.
Here, then, we individuate the perspective oered by each individual
indexical by the unit of its containing system, not by the unit of the indi-
vidual term.
Knowledge by Metaphorical Character 279

For indexicals, this systematic feature of their perspectives is, of course,

especially striking because, as Frege (1984) rst observed, there are oblig-
atory rules that direct the replacement of one indexical by another to
preserve referent, or content, under change of context. ``If someone wants
to say today what he expressed yesterday using the word `today', he will
replace this word with `yesterday'. . . . It is the same with words like `here'
and `there' '' (40). And, I would add, `I' and `you', and `today' and `to-
morrow'. These content-preserving substitutions are mandatory in all
linguistic contexts, including embedded that-clauses, under the relevant
change of extralinguistic context. So, if Jack says on Monday morning:
Jill is going up the hill
I can report later that day:
Jack said that Jill is going up the hill today
but on Tuesday I must say:
Jack said that Jill was going up the hill yesterday.
These transformations are routine. Indeed the routine, mandatory shifting
of indexicals might also be taken to show that the perspective preserved
across these substitutions, whatever it is, is also all there is to the charac-
ter-istic cognitive signicance of indexicals. For if there were additional
cognitive dierences semantically individuated at the level of the charac-
ters of the individual indexicals, we would expect them to make a dier-
ence in the substitution-behavior of the indexicals in attitudinal contexts.
The fact that they do not, that the referent/content-preserving replace-
ments are mandatory, suggests that there is no additional cognitive per-
spectival or character-istic information semantically associated with the
indexicals at the level of their characters. But, in any case, in adopting the
use of an indexical, the speaker commits himself to its entire coordinate
system, which must be systematically changed with shifts in context.
Something similar is at work with metaphor, although there is nothing
nearly as perspicuous as the perspective determined by the indexical sys-
tem. Metaphor networkswhether they are the schemes of exemplica-
tion or inductive networks or even thematic relationsare not governed
by routine, let alone mandatory, transformations of terms to accommo-
date changes of context. (Note, on the one hand, that in this respect
metaphors are more like complete demonstratives, for which there are
also no obligatory, routine rules. On the other hand, for the thematic
networks, we can predict required thematic arguments for verbs but not
280 Chapter 7

unique candidates to discharge those roles.) Yet the perspective furnished

by a metaphor network shows itself in the many ways that it guides,
directs, even commits us to go on metaphorically, to extend and elaborate
a given metaphor. When we adopt a metaphor, we adoptor inheritits
respective scheme(s) and network(s). If we change the relevant feature of
context for a metaphorthe context set of presuppositions we associate
with the vehicle of the metaphorical expressionwe must change not
only that individual expression but its whole family to preserve the con-
tent. Although you could obviously read each of the (italicized) expres-
sions in Lady Capulet's speech independently of one another and each
metaphorically, to do so would leave out something signicant about
them as metaphors: that they are all members of a single interrelated
scheme organized and weighted relative to each other. To borrow again
from Quine, the cognitive signicance specic to the metaphorical mode
of expression of a metaphor is awarded to it not as an individual term but
as a member of a corporate body (or bodies)its network or scheme.
The set of schematic, network-oriented commitments of a metaphor is a
central element in the idea of a perspective (on its content) that is con-
tributed by the metaphorical character of the metaphorical expression
`Mthat[F]'. It is no accident that, when asked what Romeo meant by
`Juliet is the sun', we reply with something like: ``Romeo means that Juliet
is the warmth of his world; that his day begins with her rising; that he can
ourish only when she shines her light on him.'' Simply to give the con-
tent of the metaphor (which we might do ``in principle'' in a literal para-
phrase) would not capture the additional perspective in which Romeo
stands toward what he says. And the only way to give that perspectival
information is by spelling out its metaphorical consequences and com-
mitments: by elaborating the place of F in its various schematic inferences
and networks, drawing out its metaphorical extensions and the other
members of its family. So, the fact that we frequently explain a metaphor
with more metaphorsnot breaking out of the metaphorical circle
should not be taken to be objectionable. Nor should the fact that we sense
the inadequacy of literal paraphrases or explications of metaphors be seen
as a problem (as many seem to have thought in the past) for the cognitive
status of metaphor. On the contrary, these facts are characteristic of the
very way in which a metaphor furnishes its cognitive perspective on, or
mode of presentation of, its content. Indeed we have now come full circle.
Not only is it not necessary (as philosophers used to think) that we be able
to literally paraphrase a metaphor in order to manifest our knowledge
of its ``meaning''; theperhaps onlyway we can manifest our under-
Knowledge by Metaphorical Character 281

standing of the mode of presentation provided by a metaphor is by elab-

orating the other metaphors in its scheme or family.
As with indexicals, a metaphor's perspective on its content is context-
oriented. Therefore, it is a function of the character of the metaphorical
expression rather than of its content (in that context). But it should be
noted that context interacts with the mode of presentation carried by the
character of the metaphor at two points. First, the scheme to which the
utterance of a given metaphor belongs, and as a member of which it
comes to have its particular mode of presentation, depends on and may
vary with the context. Second, like the character of a demonstrative, the
character of the metaphor, which carries its perspective on its content, is
nonconstant and context sensitive. Only someone occupying Romeo's
context, someone who makes (even if he does not himself believe) his
presuppositions about the sunsay, that it is the center of the universe
and that everything else revolves around itcan express what he does
with his metaphor: that Juliet is the one on whom all his actions are
focused. The mode by which the metaphor presents its content requires
the gures in Shakespeare's play, and us as interpreters, to adopt a com-
mon contextual perspective by making particular shared presuppositions.
But, to use our earlier terminology, some of these contextual presupposi-
tions are presemantic, involved in assigning its character to the metaphor,
and others are semantic: given the character, they function as its contex-
tual argument to yield content. And not unlike the problem we faced in
chapter 4, section IV in apportioning presuppositions among the semantic
and pragmatic determinants of appropriateness, here, too, it may not
always be absolutely clear how to distinguish the role of a particular
presupposition. This diculty will recur in the coming sections.

VII Metaphor and Seeing-as

In chapter 3, we saw how Frege's notion of sense reects his perspectiv-

alism. In the same vein, the (additional) cognitive signicance individ-
uated by the character of a metaphor should be conceived as an epistemic
``perspective'' on its content. Although Frege only hints at the systematic
character of his perspectivalism in his brief remarks about indexicals,
metaphorical perspective, I argued in the previous section, is deeply
rooted in the networks and schemes to which the interpreter (speaker,
user) of the metaphor commits himself in his interpretation of a meta-
phor. In this and the next two sections I want to discuss three ways in
which this network-individuated perspective corresponding to the char-
282 Chapter 7

acter of a metaphor provides both dierent information and information

of a dierent kind than its propositional content (in a context). The rst
two of these are, I want to suggest, ways to work out the often-repeated
(but rarely explicated) claim that metaphor is connected to the phenom-
enon of seeing-as: that when Romeo utters `Juliet is the sun' he sees (and
invites his interpreters to see) Juliet as the sun. The third way will return
us to our opening example of Marie and the semantic status of metaphor
in belief.
What is meant when we say that Romeo's utterance of the metaphor
`Juliet is the sun' makes us see Juliet as the sun? One thing is that Romeo
is not simply asserting the single belief that is its content in its context. By
seeing Juliet as the sun, Romeo commits himself, and asks his interpreters
to commit themselves, to a way of thinking of Juliet and her properties
that is broader than the content of a single belief. To think of Juliet as the
sun, as opposed to merely thinking that she is the sun, is to direct oneself
to think of her in terms of the scheme(s) to which the metaphor `Juliet is
the sun' (in its context) belongs.30 This commitmentto think of Juliet
and her properties in terms of a complete, complex metaphor scheme
cannot be adequately cashed out simply by enumerating each of the con-
tents of the individual beliefs that would be expressed by utterances of the
metaphors that belong to the scheme in question (even on the supposition
that we could exhaustively list all the members of the scheme). For what
that enumeration of contents would leave out is the structure of and
relations between the metaphors in the scheme, the way in which the
scheme packages those metaphorically expressed properties, including the
weighting, ordering, and organizing that expresses the speaker's and
interpreters' comparative evaluations of the properties. It is this kind of
structure that Black may have had in mind when he wrote that to use
chess vocabulary (which itself is vocabulary transferred from talk about
medieval knightly battle, a fact that curiously enough Black does not
mention) to describe a battle
will lead some aspects of the battle to be emphasized, others to be neglected, and
all to be organized in a way that would cause much more strain in other modes of
description. The chess vocabulary lters and transforms. . . . The metaphor selects
emphasizes, suppresses, and organizes features of the principal subject. . . .
Suppose we try to state the cognitive content . . . in ``plain language''. . . . The set
of literal statements so obtained will not have the same power to inform and
enlighten as the original. . . . [T]he implications, previously left for a suitable
reader to adduce for himself, with a nice feeling for their relative priorities and
degrees of importance, are now presented explicitly as though having equal
weight. (Black 1962, 4246 )
Knowledge by Metaphorical Character 283

Black goes on to say that this kind of information conveyed by a meta-

phor is sui generis. However, as Martin Davies (1982a) has persuasively
argued, the kind of structuring or restructuring of properties achieved by
a metaphor can be equally well achieved with literal language. Suppose I
have a quiet, polite neighborcall him Philbyalways responsible in
paying his bills and doing his duties, who prizes his privacy and, though
not unfriendly, keeps to himself and does not make it easy for his neigh-
bors to get to know him well; a fairly innocuous neighbor who does not
bother or disturb anyone else and who does little to attract attention.
Then one day we learn that he has been arrested as a spy for The Enemy.
Now seeing Philby as a spy, that is, thinking of him as a spy, puts all his
familiar properties in a new light. We see his dutiful, responsible, polite
behavior in light of an ulterior motivedesigned to give us a certain im-
pression. We understand dierently than we did before why he was so
guarded about his privacy and why it was so dicult to get to know him
well. We reinterpret all his eorts at anonymity and innocuousness in
terms of his newly discovered occupation. The novel piece of news that
Philby is a spy is not just a single new belief we acquire (although it is also
that); for it restructures and reorganizes all of our other (prior) beliefs
about himand will also lead us to discover, or uncover, other beliefs we
might not have noticed otherwise (say, about his passion for electronic
gimmicks and high-frequency radios).
Similarly, Davies argues, with metaphors: Seeing Juliet as the sun
restructures our complex of beliefs and attitudes about Juliet. It puts her
and her properties in a completely new light, a light that both displays her
familiar properties in ways we did not see before and reveals new prop-
erties. What is this new light? It is thinking of Juliet in the terms of the
metaphor schema to which one's utterance of `Juliet is the sun' (in that
context) commits one. As we have seen, this way of thinking of the subject
and its properties may entail revisions in our conception of the relations
between the properties expressed by the members of the schema, some-
times large-scale changes that involve redescriptions of the properties,
sometimes more subtle ones about their relative weights. Davies is cer-
tainly right that metaphor is not unique among devices that can achieve
such global, complex reorganizations and restructuring of beliefs. But a
metaphor, via its character, is especially well suited to this task because of
its place, essential to its identity as a metaphor, within a larger governing
schema: because its use carries along with it the macroscopic way of
thinking of contents that is spelled out by its network(s). As Davies also
notes (though not in these terms), this way of thinking of Juliet and her
284 Chapter 7

properties is not itself the content of any single metaphor in its context;
yet it is cognitive signicance nonethelesscognitive signicance that can
be identied only when we attend to the level of the character of the
metaphor.
Insofar as the structured schema to which a metaphor belongs creates
an organized, and thus unied, way of thinking about the properties that
are the contents of its individual members, we might say that it subsumes
them under a new complex category or concept. However, this kind of
novel categorization or ``conceptualization'' achieved by a metaphor
should be distinguished from the other way (discussed in ch. 5, sec. VI) in
which a metaphor can ``introduce'' novel properties: through the de re (in
Burge's sense) expression of properties for which we possess at the time of
utterance no context-independent, conceptualized (if you will, ``literal'')
means of expression, properties to which we can be epistemically related
only by exploiting the extralinguistic context, like the properties expressed
by predicate demonstratives. Where the knowledge by metaphor is de re
in this sense, the further cognitive signicance, or perspective, contributed
by the schema of its character, may complement itand help dene the
``bare'' property (in terms of its schematic role). But even where the indi-
vidual properties expressed by the metaphors that belong to a schema are
such that we could express them by fully conceptualized means, the eect
of the metaphor-schematic perspective will be to present them in a dier-
ent, unanticipated form (as in the Philby example) that furnishes knowl-
edge by the metaphor we did not otherwise have.
A good example of this information conveyed by the structure of a
schema beyond that of its constituent metaphors can be seen in the met-
aphorical language at work in discussions of a recent U.S. Divorce Court
case over the value of a corporate wife's work, her contribution to her
husband's career. According to Judith Dobrzynski, writing in the
International Herald Tribune (January 2526, 1997; all italics mine), the
issue is ``how much is [the corporate wife's] contribution [to her husband's
career] worth?'' And ``its resolution will be a verdict on the institution of
marriage itself and on the value of the supportive duties traditionally
known as `women's work' ''; ``Invoking economic theory, [the plainti, a
wife who turned down a $10 million settlement] is arguing that her per-
formance as a corporate wife was an investment, entitling her to half the
family fortune. . . . `Gary wanted to buy out my partnership, and I didn't
want to be bought out,' [she said, using language she learned in her role];
`It's like a hostile takeoverhe oered me a very small percentage, I said
that's not the price of the buyout.' According to Prof. Martha Fineman,
Knowledge by Metaphorical Character 285

``the important public policy issue here is: What is the nature of the mar-
ital partnership? . . . Is it an equal partnership or is a housewife a junior
partner?'' In each of these statements, there would be no diculty inter-
preting (and literally explicating) the contents of the individual (italicized)
metaphors. But the cognitive eect of the metaphor schema as a totality is
powerful in its own right. Playing on the traditional literal description of
marriage as a partnership, the metaphorical application of the contem-
porary legal/economic schemanone of whose constituents is inductively
related to the nonlegal/noneconomic termfurnishes a novel way of
thinking of the marriage institution. It selects and reorganizes the not
unfamiliar properties expressed by each of the constituent predicates to
t the legal/economic partnership model. And, to anticipate section IX,
the metaphorical schema also explains the behavior of the parties to the
divorce in ways that the received notion of marriage as (literally) a part-
nership, of course, never would.
This is also, perhaps, the cognitive signicance of Lady Capulet's use of
the `people are books' metaphor schema. It is dicult to think that the
contents of any of the individual metaphors in that passage were not
known by Juliet independently. But by reorienting her relation to Paris to
that of a reader to a book, Lady Capulet makes Juliet see her relation to
Paris as something dierent than what the contents of those metaphors
singly would have meant to her. And, again, this dierential information
can be captured only by looking at the character of the metaphor beyond
its content (in the context).

VIII Metaphor and Pictures I

There is a second signicance that Donald Davidson, among others, has

attached to the claim that a metaphor makes us see one thing as another:
the idea that metaphors are pictorial or picture-like.31 The metaphor/
picture analogy is an old and rich onemetaphors are, after all, gures
of speechand, I shall argue, it is crucial to understanding how the
character-istic perspective furnished by a metaphor is distinctive. However,
it is also primarily from this analogy that Davidson draws his well-known
denial that metaphors have propositional content. Apart from his objec-
tion that what a metaphor conveys should not be given the status of
meaning (the objection we addressed back in ch. 2), Davidson argues that
the so-called content of a metaphor is just what it makes us ``notice'' or
``see''; that what the metaphor makes us notice or see, in turn, is no dif-
ferent in kind from what a picture makes us notice or see; and that what a
286 Chapter 7

picture makes us notice or see is not content. In particular, he singles out

two features common to metaphors and pictures that legislate against
their expressing content. First, like a picture, ``there is no limit to what a
metaphor calls to our attention.''32 Second, as with a picture, when we
hear a metaphor ``what we notice or see is not, in general, propositional in
character.'' As with Wittgenstein's duck-rabbit, when I get you to see one
thing as something else, ``no proposition expresses what I have led you to
see. . . . Seeing as is not seeing that.'' Analogously, ``metaphor makes us
see one thing as another'' but this is not ``recognition of some truth or
fact'' (ibid., 263), that is, something contentful of the kind addressed by a
semantic theory.
Although I shall argue against Davidson's conclusion that metaphors
do not express metaphorically specic propositional contents, I do think
that he puts his nger on a pictorial dimension that distinguishes the
character of a metaphor from that of nonmetaphorical expressions. And
despite the fact that we have rejected (in section III) a number of argu-
ments for the endlessness of metaphorical interpretation, there is also a
kind of endlessness we have not yet examined that falls out of Davidson's
analogy between metaphors and pictures. But before turning to that
argument, I shall briey discuss another argumentbest formulated
by Richard Moran, although it expresses a widely felt intuition (which
Davidson also hints at)that metaphors do not have ``denite'' proposi-
tional content because the ``irresistible'' force of their picture-like seeing-
as violates a condition on the assertion and communication of content.33
In section IX, I'll return to Davidson's main argument.
According to Moran (1989), an assertion is characterized by the fact
that, having understood it, one is always in a position to accept or reject
it. ``There is no category of utterances that necessarily produce, when
understood, agreement or belief in what they assert'' (92). In the same
vein, ``communication involves a relation between assertion and belief,
and is always resistible. And part of what this means is that the notions of
communication and of saying require that a distinction can always be
drawn between understanding and belief '' (99). On the other hand, meta-
phorsand here Moran has in mind live, successful metaphors that
make us see (or ``frame'') one thing as (or in terms of ) anotherhave a
``compelling power'' or force that makes them ``irresistible'' (ibid.). Once
understood, they compel acceptance (as true) or rejection (as false). This
``compelling'' force or power of a metaphor, Moran further argues, is best
explained by its pictorial dimension. To understand or interpret a meta-
phor is, like viewing a picture, to see things a certain way, to see one thing
Knowledge by Metaphorical Character 287

as another, and, where that seeing-as is successful, thus to believe them to

be that way. Both successfully seeing what a picture depicts and under-
standing what a metaphor expresses have factive implications. So, when a
speaker interprets the metaphor `a is F ', if he succeeds, he sees a as an F.
But then he must also believe a to be F, not literally of course, but meta-
phorically. He must believe a to have a particular feature metaphorically
related to F. Therefore, there remains no space between understanding, or
interpreting, the metaphorical utterance and believing it; if we ``get'' the
metaphor, or understand its interpretation, we cannot but believe what it
says. But if all assertions are, by hypothesis, ``resistible''if all assertions
leave open the possibility of belief or acceptanceand what the metaphor
makes us see is not resistible, it follows that no metaphorical utterance
can express the content of an assertion.34
In reply, we can agree, to begin with, that there is a close connection
between the interpretation of a metaphor and belief. To ``get'' a meta-
phor, to understand it, is to give it an interpretation, and gifts of meta-
phorical interpretation, like other gifts of interpretation, are governed by
charity. We take speakers to mean what, in light of our beliefs, is (or is
compatible with what is) trueas well as informative, interesting, and
appropriate. Unless we have a story to explain why it was nonetheless
reasonable for the speaker to say something false by our lights, falsity
counts against a purported interpretation. As a rule, then, our interpre-
tations of metaphors express what (for all we know) is true. (Of course,
what we, or the whole community, believe to be true may turn out to be
false.) So, what a metaphor is interpreted to assert will, by and large, be
something we ought to believe.
Moran, however, seems to have something yet stronger in mind. If it is
necessary to successfully see a as an F in order to interpret the metaphor
`a is F', we must already believe that a has some particular featurethe
feature that in turn is the content of the metaphorical assertionin virtue
of which it can be seen as an F. Thus the very interpreting of the metaphor
puts us in its doxastic grip; having generated the interpretation by way of
seeing one thing as another, there is no further (open) question whether or
not it is true, to be accepted or rejected, whether, say, a has the feature in
virtue of which it can be seen as an F.
But does this show that metaphors cannot be used to make assertions? I
would still resist this conclusion. Suppose the understanding of a meta-
phor already presupposes belief in its interpretation. Contrary to Moran's
blanket claim that no class of utterances ``necessarily produce, when
understood, agreement or belief in what they assert,'' there is a venerable
288 Chapter 7

(even if criticized) tradition that there is a signicant class of utterances

with this very property: the so-called analytic (and, depending on one's
philosophical position, a priori) truths. When one grasps their meaning,
or what they assert, she already knows (hence believes) that they are (and
must be) true. It is for this reason that if someone sincerely and knowingly
denies (and, perhaps, even doubts) one of these to be true, there is often
sucient reason to doubt either her rationality or her linguistic compe-
tencewhether she truly understands the utterance.
Metaphors are not analytic truths, but the case suggests that there is
some normative relation for at least some propositions between under-
standing and belief. A second class of utterances, whose status is still
closer to that of metaphors, is that of the ``pragmatically necessary''
truths, such as `I am here now', which are true on every occasion on
which they are uttered, and which speakers are assumed to believe in vir-
tue of their knowledge of language, for example, their understanding of
the meanings of the words `I', `here', and `now' and their structural rela-
tions to each other. Whenever I utter this sentence, I know that it
expresses a truth, something that I cannot but believeeven if (for lack of
empirical knowledge of where or who I am) I do not know what propo-
sition I am asserting. With these utterances as well, there is no gap be-
tween understanding themat least understanding what they mean, if
not their truth-conditional content, or what they sayand believing them
to be true. One can resist assent on understanding them only at the cost of
casting doubt on one's rationality or linguistic competence. Here, too,
there is no further question whether the utterance is true or false, should
be believed or not, once we understand its linguistic meaning.
There remain, to be sure, dierences between metaphors and pragmat-
ically necessary truths. However, the two examples of analytic and prag-
matically necessary truths suce to show that we need not worry that we
are violating a general condition on assertion if it should turn out that,
given our practices of (charitable) interpretation, our understanding of a
metaphor, or even our knowledge that the metaphor has an interpretation
(even if we do not know precisely what it is), is accompanied by belief in
or assent to its truth. Metaphors do not ``create'' facts (or similarities), but
their interpretation may be self-verifying or self-fullling. What a meta-
phor makes us see does not make what we see, but what is there to be seen
may explain why the metaphor makes us see it. Therefore, the fact that
the seeing-as involved in generating the content of the metaphor may
presuppose belief need not impugn its assertability, nor should it be taken
to show that the content of a metaphor is nonpropositional.
Knowledge by Metaphorical Character 289

IX Metaphor and Pictures II

I now turn to Davidson's argument against the propositional status of

metaphors on the grounds that they make us see aspects or features by
way of picture-like seeing-as that is not propositional seeing-that.35 As I
said in section VIII, Davidson has drawn our attention to a feature that
distinguishes the network-oriented cognitive perspective carried by meta-
phorical characterwhich in turn enables us to make sense of the end-
lessness of metaphoralthough his argument does not show (I'll argue,
contra Davidson) that metaphors do not have propositional content. Let's
begin by putting aside two ways in which seeing-as cannot make the dif-
ference Davidson wants it to make.
First, there is an uncontroversial sense in which seeing-as is not seeing-
that. The object of seeing-that is a sentential clause, whereas the object of
seeing-as is a predicative clause (or an open sentence)the subject is seen
as having some property. This dierence, however, cannot support
Davidson's conclusion that what the metaphor (or picture) makes us no-
tice is not propositional if what we mean by ``propositional'' is not only
(in the narrow sense) a complete proposition (or closed sentence), one that
is either true or false, but also a propositional constituent like a property
that is true or false of things and that can gure in singular (or our refer-
ential) propositions.
Second, although we do say that Romeo's metaphor makes us ``see''
Juliet as the sun, this obviously cannot be taken to mean that we (liter-
ally) see her as the sun in the same way that we do (literally) see the
physical medium of a picture as what it is a picture of. There is no evi-
dence that metaphors are visual in their functioning, that they make us
visualize their content, that they evoke mental images, or that they visu-
ally depict what they say.36 Neither this nor the previous consideration,
then, supports Davidson's claim that metaphors are pictorial in a way
that would render them nonpropositional.
Nonetheless, I agree that there is a pictorial dimension to metaphor and
that the perspective it generates cannot be expressed propositionally. We
can begin to see what this perspective might be from Davidson's comment
that when the metaphor `he was burned up' was ``active, we would have
pictured re in the eyes or smoke coming out of the ears.''37 Would we
have? Well, yes and no, depending on what we mean by ``pictured.'' If we
mean `formed a mental image or picture', then perhaps some of us but not
others would have (if only because of the well-known variation among
individuals' capacities to construct mental images). This reading of the
290 Chapter 7

claim is at best inconclusive. On the other hand, what almost certainly is

the case is that when the metaphor `he was burned up' was active, or
alive, we would have also said that the subject had re in the eyes or
smoke coming out of the ears. That is, when the metaphor was alive, it
would have functioned as part of a productive network in which we
would have drawn extensions like these from the antecedent metaphor.
This relation between a metaphor and its schema, I want to argue, is the
key to its pictorial dimension and its distinctive character-istic cognitive
signicance. Let me explain.
What makes a symbol pictorial is neither that it is perceived in a visual
medium or modality, nor that it visually depicts its subject matter.38 In-
stead it is a matter of its mode of expression. Like a picture, a metaphor
displays rather than describes its content. There are a number of proposals
in the literature that attempt to articulate this distinction between pic-
tures and descriptions; the most common technical formulation is that
pictures or the contents of the visual experiences they depict are analog,
while discursive symbols and linguistic descriptions are digital. Unfortu-
nately, there is also little agreement over the theoretical explication of the
analog-digital distinction. For the sake of exposition, I'll follow Nelson
Goodman's version of the distinctionwith a little help from John
Haugeland.39
Goodman denes the distinction in terms of the nite dierentiability
of symbol schemata (sets of basic characters and rules of composition)
and symbol systems (schemata correlated with elds of reference);
Haugeland formulates it in terms of write-read computer procedures, that
is, the procedures for replicating or repeating tokens of a type. For our
purposes, the crucial issue is syntactic, but to ll out the picture I will also
lay out some of the parallel semantic notions. Three notions are basic for
us. (1) A system exhibits syntactic nite dierentiation if it is the case that
whenever a mark doesn't in fact belong to both of two characters, one can
always determine at least in principle that it doesn't belong to one or the
other. It exhibits semantic nite dierentiation if it is the case that when-
ever an object doesn't in fact comply with both of two characters, one can
always determine at least in principle that it doesn't comply with one or
the other. (2) A system exhibits syntactic density if between every two
characters a third always exists, thereby assuring that the system isn't
syntactically dierentiated. It exhibits semantic density if between every
two compliance classes a third always exists, thereby assuring that it isn't
semantically dierentiated. (3) A system is replete (as opposed to attenu-
ated) to the extent that the features of a mark pertinent to determining the
Knowledge by Metaphorical Character 291

character or characters it instantiates (if any) are many and varied. The
rst two characteristics of symbol systems are in principle all or nothing;
repleteness and attenuation are matters of degree.
With these distinctions in place we can begin to capture various ante-
cedently familiar and functionally important distinctions between symbol
systems. In particular, analog systems are syntactically and semantically
dense, since they are set up so as to denote arbitrarily small continuous
changes in one quantity by means of suciently small continuous changes
in another. Digital systems, by contrast, exhibit syntactic and semantic
nite dierentiation. Linguistic systems, once they have been suitably
disambiguated, exhibit syntactic disjointness and syntactic nite dieren-
tiation, but they lack semantic disjointness and semantic nite dierenti-
ation: arbitrarily small changes in an object can require large changes in
its accurate description, for example. Pictorial systems are highly replete
analog systems, since arbitrary and arbitrarily small changes in almost
any of a picture's pictorial properties correlate with real (typically small)
changes in what an object needs to be like for the picture in question to
depict it. (Less replete analog systems are at work in maps, diagrams, and
the like.) In general, then, except by explicit at, no feature can be ruled
out ahead of time as irrelevant to the individuation of a picture; every
dierence can potentially make a dierence.
For our purposes, the main consequence of these conditions is that the
tokens of characters in analog schemata like pictures cannot be replicated
and that tokens of digital characters can be. In the pictorial case, it is
impossible to determine to which type a given token-picture uniquely
belongs; hence it is impossible to determine unequivocally that another
token-picture is of the same type. Here, it should be emphasized, ``deter-
mine'' means know. Even if by chance we did produce a replica of a pic-
torial inscription, we would lack a theoretically possible test to ascertain
that we did. Furthermore, as Haugeland emphasizes in his formulation,
even if we might on an occasion somehow succeed in producing such a
replica, for analog schemata like pictures we lack the requisite procedures
to do so, procedures that produce absolutely perfect replicas (``positive
procedures,'' in Haugeland's terms) that can be relied on to succeed every
time.
What we need for a positive procedure is a notion of type dened for a
particular criterion or along a particular dimension (such as spelling) that
lets us disregard all other dierences along other dimensions. (If we
always had to take into account all dierences along all dimensions, no
two things could ever be ``perfect copies'' of each other.) Replicas, then,
292 Chapter 7

are anything but perfect copies of each other. Just the opposite: It is be-
cause replicas of a type are individuated by a specic criterion, such as
spelling, that it is unnecessary for two tokens to be copies to pass as rep-
licas. With pictures, on the other hand, no feature (unless ruled out by
explicit at) can be ignored or abstracted away and, because variations
are continuous or smooth, the least dierence will matter. Hence the best
available procedures produce approximations to the original, not perfect
copies. On the other hand, for lack of a single or restricted set of param-
eters or dimensions (e.g., spelling) to dene pictorial types, no two tokens
can unequivocally be said to belong to one type. Hence there are no pro-
cedures to produce either pictorial replicas or perfect copies of other
pictures.40
It follows that no description can state exactly what a picture displays,
not simply because their respective individual characters (types) are dif-
ferent but because the one is digital, the other analog, that is, because the
tokens of the rst admit replication, those of the second do not. There-
fore, we can never map members of a picture system one-to-one onto
those of a descriptional system and, if natural languages are descriptional
systems, and propositions are paradigmatically expressed by language, it
also follows that we cannot one-to-one map ``pictorial content'' onto
propositional content. Furthermore, if the cognitive value of an expres-
sion is individuated by its character and the character of an expression by
its type (and we extend characters to pictures), there always ought to be
some cognitively signicant dierence between the characters of a picture
and of a linguistic description, some loss (or gain) of information in any
correlation of pictures with descriptions. If it is indeterminate to exactly
which of indenitely many types a given picture token belongsbecause
any of its indenite number of features counts toward individuating its
typewhen we put the picture into words we ipso facto classify it arbi-
trarily under exactly one type, thereby disregarding certain of its aspects
as irrelevant for its individuation, as ``don't matters.'' Whether or not one
thinks that its analog ``character'' makes a depiction richer than a de-
scription, it is clear that any such ``translation'' would impoverish the
number of features relevant to identifying its type.41
A similar explanation applies to the pictorialness of metaphors. The
cognitive signicance of a metaphor at the level of its character in part
depends on the perspective individuated by the schema to which the
vehicle belongs. But the least dierence between two expressions
(tokens), even if both of them are interpreted metaphorically, can aect
their schematic relations. Certainly any dierence between their respec-
Knowledge by Metaphorical Character 293

tive characters will aect their schematic association, but, as we said, even
slight unpredictable dierences in context, or in the context set of pre-
suppositions, can change the constitution of the schema of a metaphor,
the other metaphors to which it is related, and hence aect its character-
istic, schema-relative cognitive signicance. Indeed not only features of
the type of the metaphor but even of its tokening may bear on its sche-
matic membership. It is in this respect that a metaphor behaves as if it
were a picture, image, or nondiscursive representation, that is, as if it were
a member of a replete dense, or analog, system of representations that
does not admit of replication, for which we possess no procedures that
produce absolutely perfect replicas. The slightest dierence between two
metaphors can make a dierence.
Let me emphasize that, when I say that the least dierence can result in
a dierence of schema, I am allowing that even the formal notion of
character may not always be individuated nely enough to capture all
relevant identity conditions for two metaphors to bear the same cognitive
signicance, insofar as that depends on their respective schemata. Con-
sider `Tully' and `Cicero' used metaphorically. These two co-referring
proper names have the same content and (because their respective char-
acters are constant) also the same character. But in some contexts one can
be used metaphorically to express a property that the other would not. In
the seventeenth century, for example (when even non-Quine-reading-
philosophers knew that `Tully' and `Cicero' were the same Roman,
namely, Marcus Tullius Cicero), one would have used `x is a Cicero' to
say metaphorically that someone is an outstanding orator but `x is a (or
my) Tully' to say that he is a defender of liberty, or to refer to him as the
author, with aection.42 For proper names like these, it would appear
that only the fact that they are lexically dierentthe fact that they are
dierent namescan mark the dierent schemata to which they belong.
Character simpliciter will not do the job.43
For the same reason, synonymy (even dened as identity of character)
is not sucient to preserve schematic association and hence the cognitive
signicance that is a function of the schema to which a given metaphor
belongs. `Sweat' and `perspiration' are about as close synonyms as any
two words (diering only in their Fregean coloring), but we clearly cannot
preserve the metaphorical interpretation of
(11) Tonya Harding is the bead of raw sweat in a eld of dainty
perspirers. (Time Magazine, Jan. 24, 1994, 51)
if we substitute the synonyms:
294 Chapter 7

*(12) Tonya Harding is the bead of raw perspiration in a eld of dainty

sweaters (sweating people).
*(13) Tonya Harding is the bead of raw perspiration in a eld of dainty
perspirers.
or
*(14) Tonya Harding is the bead of raw sweat in a eld of dainty
sweaters (sweating people).
Just as its density makes it impossible to ``put a picture into words,'' so
the fact that no featurelinguistic or extralinguisticof the utterance of
a metaphor can be discounted in determining its respective schema makes
it near impossible to express the knowledge conveyed by the character of
a metaphor except with that very metaphor. It is in this respect that (lin-
guistic) metaphors are like syntactically not nitely dierentiatedand
thus irreplicablepictures.44
On this understanding of the pictorial character of a metaphor, we can
also now make semantic sense of the idea that metaphorical interpretation
is, even within a context, ``nonnite'' or ``endless.'' The point is not that
we can go on interpreting a metaphor without end, but that we lack
knowledge of denitexed, nitisticconditions for individuating the
appropriate notion of a type for a metaphor that would exactly demarcate
the knowledge conveyed by its schema-dependent perspective. That is, we
lack knowledge of the means to express that knowledge in a nitely dif-
ferentiated symbol schemeone that is propositional. And perhaps this,
as Cavell wrote, is ``the whole truth in the view that metaphors are
unparaphrasable, that their meaning is bound up in the very words they
employ.''45
We can now appreciate what is right about Davidson's claim that what
a metaphor makes us notice is nonpropositional. Insofar as the notion of
propositional content is the idea of something that admits alternative
expressions in language, the schema-dependent cognitive signicance
individuated at the level of the character of the metaphor does not t the
bill of propositional content.46 But what Davidson fails to mention is that
this is true only of the knowledge by metaphor individuated at the level of
its character. It is not true of what the metaphor expresses in its content
(in a context). Nor, contra Davidson, does the fact that a metaphor car-
ries this increment of cognitive signicance at the level of its character
count against its having propositional content.47
Knowledge by Metaphorical Character 295

X Belief in Metaphor: Marie Again

Let's now return to Marie. What light does our account of the cognitive
signicance of metaphorical character throw on the behavior of meta-
phors in beliefs and belief-reports? To answer this, I shall rst turn back
to indexicals and then draw a moral for metaphor.
We attribute beliefs and desireson the received view: attitudes toward
propositionsamong other reasons, to explain and predict the behavior
of agents. But when we attribute beliefs using sentences that contain
demonstratives and indexicals, what plays an explanatory role is not
merely their content but also their character and, in particular, the char-
acters of their indexical and demonstrative elements. John Perry gives an
especially vivid example of this type of explanatory signicance that
attaches to demonstratives and indexicals.
I once followed a trail of sugar on a supermarket oor, pushing my cart down the
aisle on one side of the counter and back the other side, seeking the shopper with
the torn sack to tell him of the mess. But with each trip around the counter, the
trail became thicker and I seemed unable to catch up. Finally, it dawned on me: I
was the shopper I was trying to catch. (1979/1988, 83)
And at that point, Perry stopped and cleaned up his mess. Now, what led
him to change his course of action? Presumably a new belief he had
acquired. But what belief was that? Well, the belief he would have
expressed by saying to himself: ``I am the one making the mess.'' But
what belief is that? That is, if belief is prima facie a relation to a proposi-
tion, what proposition is expressed by the indexical sentence ``I am the one
making the mess'' that would also explain Perry's action? Surely he didn't
learn the proposition that the shopper with the torn sack was making a
mess; he knew that from the start. Was it, then, the proposition that John
Perry was making the mess? Nounless we add an additional but elliptic
premise that he would have, in turn, expressed to himself as `I am John
Perry'. For simply coming to believe the proposition that John Perry was
making a mess would not be enough to make him stop unless he also
believed that he himself was John Perry. And once we add the belief
expressed as `I am John Perry', we are back to a belief expressed by a
sentence that contains an indexical. Similar considerations hold for all
other candidate representations of the required belief. Unless we build
into our representation of Perry's belief that he believed that he himself
is the individual making the mess, we cannot explain why he acted as
he did. Hence, in Perry's belief, the rst person indexical `I' occurs
296 Chapter 7

essentially; it is impossible to eliminate the indexical, or the indexical

character by which he represents his belief to himself, if his belief is to
explain his behavior.
To get a better sense of the power of indexicals to explain action, con-
sider another example oered by both Kaplan (1989a) and John Perry
(1977). Suppose you (Max) and I (JS) are walking toward each other
down a dimly lit street and I see a mugger steal up behind you. I yell out
in warning:
(15) You are about to be mugged
and, given that belief, run for help. You, stunned by what is about to
happen, yell out (perhaps echoing me):
(16) I am about to be mugged
and, doing what one does with that kind of belief, mobilize yourself and
turn to face your attacker while assuming a martial arts position.
Now, both (15) and (16) express the same singular proposition:
(17) hMax; About to be muggedi.
Our dierent actions (I run for help, you defend yourself ) cannot, then,
be explained by the contents of our utterances. And if our respective
utterances correctly express the contents of our respective beliefs, it fol-
lows that our dierent actions also cannot be explained by the contents of
our beliefs, namely, (17). Instead, the actions seem to correlate with the
characters of the sentences we uttered. Indeed, if I had the same martial
arts training, I would act the same way you did if I were the one about to
be mugged, and you (presumably) would act the same way I did were you
the warnerthough the content of my utterance of (16) and your utter-
ance of (15) would not be (17) but (18):
(18) hJS; About to be muggedi.
Same content, dierent characters, dierent actions; dierent contents,
same characters, same actions.
Perry and Kaplan conclude from examples like this that belief cannot
simply be a two-place relation between a believer and proposition(-al
content)be it singular (as in the New theory) or fully conceptualized (as
in the Old theory). Instead we need a three-place relation between believ-
ers, propositional contents, and ``ways of thinking'' of contents. Perry and
Kaplan dier in their specic formulations for this third element, but for
both of them there is an additional kind of cognitive element that is indi-
Knowledge by Metaphorical Character 297

viduated semantically, either by the characters or sentence types of action-

explaining beliefs containing indexicals.
It would be helpful to distinguish two separate claims in the conclusion
of this argument. The rst is that it is necessary to acknowledge ``ways of
thinking'' of the contents of beliefs in addition to the contents themselves.
The second thesis is that arbitrary ``ways of thinking'' can be individuated
by indexical characters or, more generally, by the sentence-types of the
utterances accepted by the speaker. I shall grant the rst, but the second is
more problematic. What complicates matters is that a believer may have
more than one way of thinking of a given content, even given one
sentence-type or given one character by means of which he expresses the
content.48 This is obvious, as we mentioned earlier, with expressions like
proper names whose characters are constant. But, even with nonconstant
characters like that of `I', we don't have to make ourselves into Dr.
Jekylls and Mr. Hydes to acknowledge that most of us have substantially
dierent ways of thinking of ourselves on dierent occasionsgiven our
other beliefs, moods, purposes, and so onand that dierent individuals
can have dierent ways of thinking of themselves even though we all use
the rst-person indexical `I'. But if there can be this kind of variation
from context to context among our ways of thinking of things, we cannot
in general systematically individuate ways of thinking by characters or
linguistic types of arbitrary expressions or sentences.49
A similar story can be told for metaphors in belief. The metaphorical
characters by which beliefs are expressed, and not merely their proposi-
tional contents, may play an analogous essential role in the explanation of
their agents' behavior. Recall Marie's utterance (repeated here):
(19) I won't swallow that [referring to her mother's interdiction].
Its content in its context of utterance is expressed by (20) (repeated here):
(20) Marie won't obey her mother's interdiction.
But that expression of its content does not enable us to explain Marie's
anorexic behavior. Now, the character of the sentence she in fact uttered
is not, precisely speaking, that of (19) but that of (21):
(21) I won't Mthat[`swallow'] that [referring to her mother's
interdiction],
which contains a metaphorical expression whose character, in turn,
determines the corresponding propositional content, expressed by (20), in
its context. Moreover, (21) contains the information that is relevant to
298 Chapter 7

explaining Marie's behavior: the information conveyed by the character

of the metaphorical expression Mthat[`swallow'] by which she represents
its content (in that context) to herself. Therefore, were we to report what
Marie believes in order to explain her eating disorder, we must report the
content of her utterance under its metaphorical character. This is to say
two things. First, this character-istic information is information that is not
part of the propositional content (in any particular context) of the sen-
tence whose character it is, and, for that reason, it will not be expressed in
any statement of that content. Second, it does not follow from the rst
point alone that this character-istic information is necessarily information
of a ``dierent kind'' from that carried by content, information that could
not be propositionally expressed (say, as the content of some other sen-
tence). However, it may be of a dierent kind. As we just saw in the case
of indexicals, it may not be possible to individuate all ways of thinking
relevant to the explanation of action that are expressed by a metaphor by
its character. The perspective conveyed by a metaphorical character is
relative to the schema to which its vehicle belongsand that perspective,
as argued in the previous section, lends a pictorial dimension to the meta-
phor. So, to the extent to which its perspective enters into the explanatory
role of the metaphor, one character may be associated with dierent
action-explanatory ``ways of thinking'' for one metaphorical expression
on dierent occasions, and, furthermore, those ways of thinking may
contain nonpropositional pictorial elements.
Nonetheless, like Perry's example of an ``essential'' indexical, the case
of Marie is an example of an ``essential metaphor.'' It is the metaphorical
character of the expression she used (or the character of the metaphorical
expression), her way of thinking of the information contained in the con-
tent of the metaphor rather than the content itself, that is essential to
explaining her action. Much more, of course, remains to be said about the
role of metaphors in belief and other attitudinal reports and in indirect
discourse. In particular, I am not claiming that it is never sucient to re-
port the content of a subject's utterance of a metaphor without preserving
its metaphorical character. For it is often the case that belief ascriptions
are vehicles for reporting only the contents of a subject's beliefs rather
than the manner in which he holds them or how he thinks about their
content, and belief reports based on utterances containing metaphors are
no dierent. However, our discussion does show that a general account of
metaphor in belief reports should take into account not only their propo-
sitional contents but also ways of thinking of those contents, both indi-
viduals and properties. Insofar as we acknowledge that the character of
Knowledge by Metaphorical Character 299

an indexical possesses its own explanatory role and cognitive signicance

that must be incorporated in a complete theory of belief, we should do no
less for metaphorical character.

XI The Moral of the Story

We are now in a better position to understand both the problem of

knowledge by metaphor and the errors in some of its traditional and more
recent diagnoses. Decoration theorists failed to recognize, and therefore
disregarded, what is cognitively signicant about metaphorical character.
In the same way, those recent philosophers who attempt to focus the
question of metaphorical interpretation solely on propositional content
fail to do justice to what is specic to metaphor by relegating it to vacuous
explanatory factors like ``the special ways that metaphorical utterances
have of calling other things to mind.''50 Autonomy theorists, on the other
hand, have recognized what is cognitively signicant about metaphorical
character but, lumping it together with propositional content, have taken
it all to be sui generis and mysterious. Philosophers like Davidson, who
could be said to focus entirely on the signicance specic to metaphorical
character, are thereby (mis)led to deny that metaphors have any proposi-
tional content. I have tried to steer a middle course between the omissions
of the decoration theorists and the excesses of the autonomy theorists.
By identifying the cognitive signicance specic to a metaphor with
its character and distinguishing it from its propositional content, we can
acknowledge that there is a genuine semantic and cognitive dierence
between the metaphorical and the literal without making it occult. What
distinguishes metaphors is their context-sensitive character-istic perspec-
tive. By subsuming metaphors within the general category of expressions
of nonconstant character, including demonstratives and indexicals, we
can both recognize what makes them dierent and begin to understand it.
Chapter 8
From the Metaphorical to
the Literal

In closing, I want to take up two questions raised by my account that

touch on the boundaries of metaphorical interpretation: the range of
symbols that can be interpreted metaphorically and the range of inter-
pretations that are not metaphorical. Is metaphorical interpretation
limited to natural language? And what is a nonmetaphorical, or literal,
interpretation of language? I'll touch on the rst question briey; the sec-
ond will occupy most of the chapter although I shall not answer it con-
clusively. Nevertheless, having come this far, we are in a much better
position to articulate both questions. That alone is a small step of
progress.

I Nonlinguistic Metaphors

We often hear it said that a picture (say, of Napoleon in a Roman toga) is

metaphorical, or that a dance gesture is a metaphor for an emotion. Pablo
Picasso called his sculptures ``visual metaphors.'' Anthropologists tell us
that various social interactions and religious rituals should be explained
as metaphors. George Lako suggests that a depiction of smoke or steam
coming out of a cartoon character's ears is a metaphorical depiction of
anger, and a character falling on his face, a pictorial metaphor for social
clumsiness.1 A video blurb for Goddard's movie ``Weekend'' describes ``a
weekend trip from Paris to Normandy as a shattering metaphor for the
decline of the West.''
How should we understand these claims of metaphoricity? Are non-
linguistic symbols, objects, and events the sorts of things that can be
interpreted or used metaphorically (or, for that matter, literally)? I have
argued that an essential element of metaphorical competence is semantic,
the same semantic competence that underlies speakers' abilities to inter-
pret demonstratives and indexicals. But the semantic competence underly-
302 Chapter 8

ing the ability to use demonstratives and indexicals is arguably task-specic

to language. Of course, in addition to semantic knowledge, there is also
an extralinguistic factor in the interpretation of demonstratives and
indexicals: the contribution made, for indexicals, by their contextual
parameters and, for demonstratives, by their completing demonstrations,
either presentations or pointings. Likewise, we have acknowledged the
essential role of its extralinguistic contextual parameter, the context set of
presuppositions, in metaphorical interpretation. Nonetheless, if a neces-
sary component of metaphorical interpretation is task-specic linguistic,
or semantic, knowledge, it should follow that only objects of language can
themselves be metaphors. And if that is the case, all these nonlinguistic
symbols would be at best metaphorically, but not literally, metaphorical.
Or our description of these nonlinguistic symbols, or symbolic interpre-
tations, as metaphor would be parasitic on our talk of linguistic meta-
phors; that is, without a linguistic gloss on these symbols, they would not
admit metaphorical interpretation.
The issue is not terminological, and it also cannot be answered (simply)
by appeal to ordinary language. The question concerns the mechanism
that is essential to and distinctive of metaphor. Are metaphors distin-
guished by the sources of the contextual presuppositions on which their
interpretations dependthe extralinguistic skills employed in the recog-
nition of similarities, exemplication relations, or salience of other kinds
(including the prominence attached to stereotypical features or normal
notions)? Likewise, maybe it is their extralinguistic demonstrations that
distinguish demonstratives. Or are metaphors (or the class of M-gures)
and demonstratives really distinguished by their respective kinds of seman-
tic context-dependence involving specic constraints on their possible
interpretations?
My approach has emphasized the specically semantic, hence linguistic
knowledge employed in metaphorical interpretation. This focus should be
contrasted with Nelson Goodman's, for whom there is a set of symbolic
skills that cuts across all symbol systems (natural languages, pictures,
musical notation, sketches, etc.) and modalities (the visual, auditory, etc.),
one mode of which is metaphor.2 Although Goodman, like myself, argues
that it is not enough to look at the function or eect of a metaphor, that
there must be some mechanism specic to metaphor that accounts for its
eects, we dier over whether that mechanism is language-specic or not.
Elsewhere, in Stern (1997), I have argued that Goodman's own proposed
mechanismtransferwill not work for pictures and other symbols that
do not admit replication of tokens of one type. But in part our dierent
From the Metaphorical to the Literal 303

views depend on dierent pretheoretical notions of what is essential or

distinctive to metaphor and our accounts of the nonmetaphorical or lit-
eral. Since a full answer to the rst question therefore rests on a better
understanding of the literal, let me turn now to the boundary between the
metaphorical and literal.

II Historical and Contemporary Notions of the Literal

How, in light of my account, should we demarcate, within natural lan-

guage, the literal and metaphorical? In chapter 1, I proposed, as a work-
ing hypothesis, that the literal meaning of simple expressions should be
whatever our best semantic theory decides is their semantic interpretation
and, for complex expressions and sentences, the rule-by-rule composition
of the literal meanings of their constituents. In constructing my theory of
metaphor I have also tried to avoid loading more weight on the literal
than can be reasonably borne by the semantic. Still, a nagging problem
that repeatedly surfaces in discussions of metaphor is its relation to the
literal. Some (including deconstructionists and lit crit theorists) deny that
there is a distinction to be drawn between the two, period. Others (e.g.,
George Lako and his school) acknowledge a distinction but claim that,
despite the ``scientic philosopher's'' prejudice for a purely literal lan-
guage, all or the most important part of natural language is metaphorical.
Still others (such as Davidson) claim that all there is is the literal, at least
if we are talking about meaning; the metaphorical is just a class of eects
that could equally well be eects of other kinds of causal antecedents.
And still others (such as Ted Cohen 1997) insist that the distinction is so
self-evident that it needs no defense. As I said back in chapter 1, I assume
that there is a distinction, if only because it is close to an analytic truth
about metaphor that it ``depends'' on the literal. Whatever that turns out
to mean at the end of the day, dependence requires a distinction.
The question concerns the dierence in what the interpreter knows
when she knows the literal and metaphorical interpretations of an ex-
pression or sentence. I am not concerned with how she processes this
knowledge or the psychology of language comprehension or production.
Many psychologists in recent years have argued that there is no dierence
in length of time or speed of processing in speaker-hearers' production or
comprehension of literal and metaphorical language. Be that as it may,
the fact that speakers process literal and metaphorical interpretations
in the same way or at the same speed is compatible with the existence of
two dierent kinds of interpretation corresponding to the literal and
metaphorical.3
304 Chapter 8

Unlike metaphor, the literal has received relatively little sustained study
of its own. Let me begin by bracketing a number of dierent uses of
the words ``literal'' and ``literally'' that are irrelevant for our theoretical
purposes.
1. When we talk about what a sentence ``literally'' means, sometimes we
intend what it precisely or univocally or specically means. But the literal
need be no more precise, univocal, or specic than the metaphorical.
When Romeo calls Juliet `the sun' (in his respective context), that is pre-
cisely, specically, and univocally what he means; I cannot imagine a
more precise, specic, or univocal way of describing Juliet than that.
2. At one or another time in history, philosophers have used the ``literal''
to refer to the empirical or factual. This use reects a particular theory of
meaning, vericationist or empiricist, that many of us no longer share. In
any case, our use of the term nowadays does not and need not carry that
baggage.
3. Some writers take ``literally'' to mean ``actually,'' and then use this
assumption to argue that metaphors, not being literally true, are also not
(indeed cannot be) actually true. And because the truth with which we
are typically concerned is the actual truth of our utterances, they also
conclude that metaphors, not being actually true, are not true, period.
Therefore, metaphors cannot be asserted; for assertions are uses of sen-
tences in which the speaker represents himself as intending to speak a
truth.4 This understanding of ``literal'' as ``actual'' rests on a confusion.
What is ``actually'' true is simply a proposition that is true in the actual
world, namely, the circumstances of the context in which the utterance is
performed. Contraries of the actual are the merely possible and the con-
trafactual.5 The distinction between the metaphorical and the literal, on
the other hand, is a distinction between two kinds of interpretations or
uses of language, not between kinds of truth, or between the circum-
stances in which what is said is true or false. Metaphors no less than lit-
eral utterances of sentences can therefore be actually truejust in case
their interpretation, or what they are used to say, is true in the circum-
stances in which they are uttered.6 In any case, actual is not what we
mean by ``literal.''7
4. What is often said to be literal is not to be interpreted literally. As
Vincent Canby once wrote:
How many movies have you seen that literally froze your blood, or literally left
you breathless, or literally drove you up the wall? If you can name one, you are
dead or have a serious medical problem, or can defy gravity. (New York Times,
Jan. 14, 1979; 17)
From the Metaphorical to the Literal 305

A word about the history of the notion of literal meaning may throw
some light on our problem. The origins of the ``literal''``of or pertaining
to the letter'' (O.E.D.)are obscure, but the dominant original context
for the use of literal meaningsensus literaluswas medieval (Christian)
scriptural interpretation.8 In On Christian Doctrine (1958), Augustine
distinguishes two classes of signifying entities: words and extralinguistic
things (persons, events, objects, actions, places, positions, times). That is,
apart from linguistic signs, there are ``natural signs,'' (extralinguistic)
things that by their nature signify, much as smoke is a sign of re. Among
these, all the things signied by the words of Scripture are, in turn, signs
of a ``higher'' or ``deeper'' spiritual order. And among these latter signi-
cations, three kinds are usually distinguished: the moral (something sig-
nied about how one ought to act), the allegorical (something signied
about the Church), and the anagogical (something signied concerning
God, the saints, angels, and other heavenly beings). These kinds of signi-
cations of things are akin to the property or concept exemplied, or re-
ferred to, by an object, as I argued in chapter 5. None of these is directly,
however, a meaning of a word.
Among the meanings of words, the medievals sometimes distinguish the
proper or analogical (of which a word can have more than one, in which
case it is ambiguous) and the improper. Among the latter, a leading kind
is metaphor. On other occasions, they rst classify the meanings of words
into three functionshistory, fable, and argumentand then, as part
of the text's historical role according to which its words set out things as
they actually happened, they distinguish the various analogical, or proper,
``modes'' of signication from the metaphorical mode. In either case,
metaphors (e.g., `the lion of Judah' referring to Christ) are understood as
comparisons between the subject and the thing signied by the predicate
(`lion'); hence they are meanings or signications of words rather than of
things. On the other hand, in addition to their proper and metaphorical
meanings, which words possess directly and by which they signify things,
indirectly the words also inherit the meanings or signicances of the things
they signify. So the meaning of the word is never the whole of its ``mean-
ing'' or signicance; indeed it is often the least important signication.
Now the term ``literal'' is sometimes used in this medieval literature
interchangeably with the meaning of a word, as opposed to a meaning of
a thing. In that sense the literal ``includes'' the metaphorical. Sometimes it
is used interchangeably with ``historical,'' in which use it signies a spe-
cic function of the meanings of words: to set down what actually oc-
curred. Here too the literal or historical use of language can work either
properly or metaphorically, subsuming both. And sometimes the ``literal''
306 Chapter 8

means specically the ``proper'' use of a word as opposed to a metaphor-

ical or metonymic use, which is ``improper.'' What makes a use ``proper''?
During the earlier Middle Ages, we often nd the literal identied with
the historical because it is at that level that the text tells a story. However,
in the rst half of the twelfth century, it has been argued (e.g., by Evans
[1984], 68), a sharper distinction begins to emerge between the historical,
which is concerned with narrative, and the literal, whose concern is the
relation between words and the things they signify. A literal (use of a)
word is now said to be a (use of the) word that signies what it does ``in
the rst place'' ( primo loco). This seems to be one sense in which a mean-
ing is properwhich, in turn, is contrasted with gurative meanings,
which are transferred and thus not in their rst place.9 But there may
also be a second sense of ``proper'' at work here. The Latin term propria
for ``proper'' also has the sense of ``usual'' or ``regular,'' that is, according
to a rule, not in a normative sense but in the sense of conforming to a
regularity.10 Thus the literal would simply be common usage of the word.
I have recounted this rough story about the history of the literal in
order to give a sense of the many pieces of baggage the term carries. The
literal is the meaning of words as opposed to things; it is the meaning
whose primary function is descriptiveto tell a narrative; it is a primary
as opposed to derived, or transferred, meaning; and it has the primacy of
the proper, the rule-governed or perhaps just the regularly used. These
features do not combine into one neat package, and insofar as the term is
used sometimes with one, sometimes with another trace of its history in
mind, we should not expect sharp uniformity of usage or meaning. But
with this background, I now want to turn to some implications of my
account for our current understanding of the notion of the literal. Let me
begin with three consequences of my approach for any analysis.
To begin with, my theory rules out certain standard ways of drawing
the literal/nonliteral-metaphorical distinction. Since I take metaphorical
competence, that is, the speaker's knowledge of the metaphorical charac-
ter of expressions (or the characters of metaphorical expressions), to be
semantic, we obviously cannot distinguish the metaphorical from the lit-
eral by saying that the latter is the object of semantic competence whereas
the former is not (but is instead pragmatic). Likewise, we cannot subsume
the literal/metaphorical distinction under the distinction between sentence
and speaker's (or utterance) meaning.
By the same token, we cannot distinguish the literal from the meta-
phorical meaning of a sentence (or of any other complex expression) by
means of compositionality. On that characterization, a literal sentential
From the Metaphorical to the Literal 307

meaning would be one that is uniquely composed from, or functionally

determined by, the meanings of its constituent expressions and their
syntax, whereas a metaphorical interpretation would be achieved non-
compositionally, top-down rather than bottom-up, subject to all sorts of
noncompositional contextual contributions. The diculty with this view
of the distinction has to do primarily with its statement of the principle of
compositionality. Unless we are told more about the kinds of meanings,
the syntax, and the sense in which the whole is a ``function'' of its parts,
the principle is tautologous or empty.11 On the one hand, it is true that
the content of a sentence at least one constituent of which is interpreted
metaphorically is not compositional, but that is equally true of the con-
tent of (utterances of ) sentences containing indexicals and demonstra-
tives, utterances of sentences like `I am happy'. What follows from that
observation is not that (an utterance of ) such a sentence is not meant
literally, but that, if compositionality is a criterion of the literal, it is not
a principle that applies (only) at the level of content. On the other hand,
if it is the character of a sentence to which the principle of composition-
ality applies, then sentences containing metaphors might also be compo-
sitional. Metaphorical expressions, like dthat-descriptions, are (at least
lexically) compositional. To be sure, there remains the problem of indi-
viduating the characters of eternal co-intensional expressions, but that,
again, is a problem for both the metaphorical and the literal. In either
case, compositionality will not distinguish the literal and metaphorical.
A second consequence of my account is that, strictly speaking, there are
no literal or metaphorical expressions per se (except as terms of art); there
are only literal and metaphorical interpretations of expressions. The same
holds, as we'll see in a minute, for the distinction between dead and live
metaphors: It is their interpretations rather than the metaphorically
interpreted expressions themselves that are dead or alive. Thus consider
the many cases where a dead metaphor is brought back to life by verbal
resuscitation: where `hot as hell' gets new life as `hot as the hinges as hell',
and `full of wind' becomes, in a poem of Yeats, `an old bellows full of
angry wind'. We can also sometimes witness a metaphorical reincarna-
tion take place by reuniting a lost, isolated metaphor with its family: by
placing a given expression, with an interpretation that we may have
learned to assign to it primitively, and not metaphorically, in the context
of the family or network of metaphors to which it historically belonged.
(I would suggest this as an explanation of the powerful impact on us of
many of Lako and Johnson's examples of systems of ``conventional
metaphors.'') By recovering the network from which an expression
308 Chapter 8

acquired its present, even if no longer metaphorical, interpretation, we

can bring its metaphoricity back to life.
The third implication follows from my characterization of metaphor
as a special kind of context-dependent expression, an expression whose
character is sensitive to its context set of presuppositions. This might be
taken to suggest that the literal should be characterized as context-
independent interpretation. And I shall argue in section IV that this is
indeed one dimension of the literal. However, it might also be objected, as
John Searle (1983) has, that all sentences, including the literal, are really
context-dependent; that sentences with the same literal meaning deter-
mine dierent truth-conditions in dierent contexts. In my terminology,
that would be to say that one literal meaning determines dierent truth-
conditions relative to dierent sets of presuppositions or beliefs; in
Searle's terminology, it means relative to dierent Backgrounds (where a
Background is a nonrepresentational mental state, to avoid what he
thinks would otherwise be an innite regress). Searle's argument for this
claim does not work, however. Let me explain.
Searle (1983) begins by asserting that ``it seems clear to'' him (145.)
that the word `open' has the same literal meaning in all of the following
sentences.
(1a) John opened the door.
(b) Mary opened her eyes.
(c) John opened his book to p. 37.
(d) The surgeon opened the wound.
But Searle does not tell us what he means by ``literal meaning'' or how he
individuates literal meanings, the kinds of considerations that would jus-
tify his unqualied, condent assertion that all of these sentences have the
same literal meaning. His only argument is that, if we allow that these
tokens of `open' have dierent literal meanings, ``we would soon be forced
to the conclusion that the word `open' is indenitely or perhaps even in-
nitely ambiguous since we can continue these examples; and indenite
ambiguity seems an absurd result'' (146). I agree: indenite or innite
ambiguity would be an absurd result. However, it is also absurd to jump
from four- or ve-way ambiguity to indenite or innite ambiguity. Searle
has given us no reason to think that if this expression has a xed even if
multiple number of dierent senses that it has an indenite, let alone in-
nite, number. Furthermore, Searle's claim that each of these sentences
has dierent semantic-conditions, that is, truth-conditions, is, given the
From the Metaphorical to the Literal 309

absence in them of explicit context-dependent expressions (e.g., indexicals

or demonstratives), as strong evidence as there could be that they do have
dierent meanings. What would instead have been a good argument for
the context-dependence of literal meaning in an objectionable sense would
have been evidence that any one of those sentences, with its literal mean-
ing held constant, has dierent semantic or truth-conditions in dierent
actual contexts. But Searle cites no evidence of that kind.12
What Searle's examples do illustrate nicely is (in the terminology
introduced in ch. 2) the presemantic context-dependence of all, including
literal, language. It is only given all kinds of background that we assign a
particular type, with its meaning (or character), to a particular token. But
that leaves open the possibility of characterizing the literal as a kind of
semantic context-independence.13 Yet that conclusion also cannot be the
whole story if only because of the semantic context-dependence of index-
icals and demonstratives. But it is a step in the right direction.
To move further in this last direction, I want to take a closer look at the
way metaphors become semantically context-independent and thereby lit-
eral. This process is how metaphors have standardly been said to die
keeping in mind that dead metaphors are also often claimed to be the stu
from which the literal grows. If we can understand this organic process
better, we may also be able to get a better grip on the literal.

III Dead Metaphors

The title ``dead metaphor,'' as it is used in the literature, covers a variety

of examples. First I'll bracket two of these, after which I'll turn to three
kinds of genuinely metaphorical dead metaphors.
First, there are expressions whose (once) literal interpretation is now
dead and long forgotten; only their (once) metaphorical interpretation
survives, although at present it is not even recognized that it is (or was)
metaphorical. For example: `plastron' now refers to the lower shell of a
turtle but originally denoted the breastplate of a suit of armor. Appar-
ently the metaphorical interpretation drew on a perceived (functional)
analogy between the turtle-shell and breastplate. Likewise, `cosmos' orig-
inally referred to a woman's headdress, was later metaphorically applied
to the ranks of an army, and only nally to the order of the universe.14
Thus the original application of `plastron' to the turtle-shell or `cosmos' to
the universe was metaphorical, although our use of the predicates now
depends on no presuppositions about, say, armor or headpieces, appli-
cations that would now be archaic. In short, the present character of
310 Chapter 8

interpretation of these predicates has nothing metaphorical about it. To

call them ``metaphors'' is only to recall their historical genesis, and to call
them ``dead metaphors'' is just to comment on their ancestry.
A second group of dead metaphors that are only debatedly metaphors
are so-called double-function terms (physical or sensory terms like `hard',
`deep', and `bright' that are applied to psychological states) and synaes-
thetic adjectives (predicates taken from one sensory modality that express
features applicable to a second sensory modality). These ``metaphors''
follow relatively regular paths of development both in individuals' acqui-
sition of language and in the growth of languages. But there is an impor-
tant disanalogy between the phylogenetic and ontogenetic cases. In the
phylogenetic case, there is clear ``transfer'' from one sensory domain to
another and, as Williams (1976) shows, even apparent directions to the
paths of transfer. Ontogenetically it is a dierent story. Although the child
may rst acquire the physical/sensory meaning of the expression (or its
application to a home sensory domain) and the psychological (or second
sensory domain) meaning later, the latter may be acquired independently
of the former. Despite the prima facie hypothesis that the term is extended
on the basis of a similarity between the two extensions, in fact similar-
ityor any other relation that might provide a ``common core'' mean-
ingseems to play little role in the child's process of acquisition.15
Initially these double-function or synaesthetic predicates are homonyms
for the child; hence ontogenetically it would be more plausible to count
each meaning as primitive rather than derived. Only at a considerably
later stage does she discover their phylogenetic or diachronic relation
that at some previous time in the history of the language one was literal,
the other metaphorical. But what she then discovers is something about
the history of the language, rather than a fact about her own inter-
pretations. Of course, the discovery might have the eect of enriching her
subsequent interpretations, bringing her presuppositions back into the
process of interpretation and at that later stage making it metaphorical.
But it is necessary to distinguish the individual's acquisition of the ex-
pression and its emergence in the growth of language. Perhaps we should
distinguish whether an expression is a (dead) metaphor in the language
from its status for the individual. Here, in short, ontogeny does not reca-
pitulate phylogeny.16
The two kinds of dead metaphors I have just discussed are arguably not
metaphors. What, then, are examples of genuine dead metaphors, that is,
interpretations that are clearly metaphorical but dead or dying? Again,
the easy part is saying what they are not. Dead metaphors should not
From the Metaphorical to the Literal 311

be confused with metaphors that are merely trite, cliched, or tired. The
deadness of a metaphor is also not to be measured by its frequency of
utterance. Some metaphors are as alive and as novel on their hundredth
utterance as they were on their rst; witness `Juliet is the sun'. But if not
by being repeated to death, how does a metaphor die? And how does the
content of an expression, once determined metaphorically, become its (or
a literal) meaning? Before making a rst stab at an answer, let me em-
phasize a point I made earlier: I am not asking the diachronic question,
How in the development and growth of a language did expressions that
were metaphorical become literal? My question is synchronic and indi-
vidualistic: What determines whether the interpretation of an expression
(uttered by a speaker in a community at a time) is metaphorical, even
dead metaphorical, or literal?
A rst stab at an account of dead metaphor might be in terms of how
the metaphorical interpretation is learned, acquired, or assigned. One idea
in the literature is that a metaphorical interpretation of an expression F is
dead just in case its acquisition does not ``depend'' on the literal meaning
of F, or if (and perhaps only if ) it is learned ``independently'' of its literal
meaning.17 But the diculty with this proposal is that we cannot severe
the connection between the metaphorical and the literal except at the cost
of rendering the interpretation not just dead but non-metaphorical. In-
stead of characterizing metaphorical-literal dependence as a relation that
either does or does not hold, let's distinguish both degrees of dependence
on (contextual presuppositions related to) the literal (vehicle) and the de-
gree to which the metaphorical interpretation depends on presuppositions
specic to its actual context of utterance. I propose that a metaphorical
interpretation of an expression F in a context c is alive to the degree to
which that interpretation in c depends on presuppositions associated with
F specic to c. As the interpretation of the metaphor ceases to be sensitive
to or dependent on presuppositions specic to its actual context of utter-
ance, the metaphor dies. And as the interpretation ceases to be context
dependent at all, it becomes literal.
There are at least three kinds of examples of metaphors that don't meet
(to dierent degrees) these conditions for metaphorical aliveness; their
interpretation is metaphorical yet dead or dying. They also suer from
dierent causes of death. If we can better understand what makes each of
them a dead metaphor, we can get a better grasp on the various elements
in our pretheoretical notion of the literal.
1. Routinized metaphorical interpretations: Some expressions F inter-
preted metaphorically always turn out to have the same interpretation, no
312 Chapter 8

matter what their specic context of utterance. Examples of these inter-

pretations include the ``default'' interpretations of metaphors that exploit
the normal notion or stereotypical features associated with F. The inter-
pretation is not (uniquely) determined by the linguistic meaning of F. The
interpreter computes extralinguistic presuppositions about features asso-
ciated with F to yield a content for F in the context. Butand here is the
distinguishing markthe interpreter brings the same presuppositions to
bear on the interpretation of F in each context, thereby yielding a con-
stant content as its metaphorical interpretation. Or perhaps, although she
knows that P is a metaphorical interpretation of F rather than its literal
meaning, the interpreter nonetheless routinely assigns P as the meta-
phorical interpretation without actually ``computing'' it as the value of
Mthat[F] in c. These metaphors, as it were, carry their context on their
sleeve. Since the metaphor has the same interpretation (or we might say:
to the degree to which the metaphor has the same interpretation) when-
ever it is uttered, drawing on the same set of presuppositions, its inter-
pretation is not dependent on presuppositions that are specic to its actual
context of utterance. It is (on its way to becoming) de facto dead. De jure,
however, it is alive. Were we asked, for example, to justify the interpre-
tation assigned to the expression, we would reconstruct the content as the
value of a nonconstant metaphorical character applied to a context set of
presuppositions.18
2. Root-inized metaphorical interpretations:19 Think of a metaphor
found in either a foreign or ancient literary text, something whose con-
text of origin (utterance, literary production) is far removed from the
interpreter's actual contextwhere the relevant contextual feature is, of
course, the context set of presuppositions. Suppose we know as inter-
preters that the expression was intended by its author to be interpreted
metaphorically. Suppose we can give an acceptable interpretation to the
inscription only by typing it as a metaphor. But suppose, too, that it is
only by recovering the particular presuppositions that obtained in its
context of origin that we can give the metaphor an acceptable interpreta-
tion (on whatever criteria of acceptability we choose). That is, suppose
there is absolutely no other set of presuppositions of our ownin our
actual contextby which we can acceptably interpret the metaphor. We
say to ourselves: ``Unless we go along with the beliefs and presupposi-
tions of its authorstheir idiosyncratic beliefsthere is no making sense
of this metaphor.'' We can, to be sure, ``recover'' the original presupposi-
tions, and it is by applying the metaphorical character of the expression to
those presuppositions that we yield its content. Hence there is no question
From the Metaphorical to the Literal 313

that the expression can be interpreted metaphorically. However, because

of the particular dependence of the metaphor on those specic pre-
suppositions, its acceptable interpretation is ``rooted'' in a particular
context, here its context of origin. To interpret the metaphor, what we do
is attempt to ``re-create'' or, for the sake of interpretation, put ourselves
into that original context. But to the degree to which the expression, not-
withstanding its acceptable interpretation according to those nonactual
presuppositions, resists alternative interpretationsto the degree to which
we nd it impossible to assign any acceptable interpretation (by whatever
standards) to the metaphor according to our presuppositions (or the pre-
suppositions of any other counterfactual context)to that extent the
metaphor is dead. Even apart from the fact that the alternative interpre-
tation would not be, we know, what its original author intended, a live
metaphor is one that has the potential for dierent interpretations in dif-
ferent contexts. Metaphors (metaphorical expression types) that don't
admit that power of interpretationbecause there aren't available alter-
native sets of presuppositions given the choice of metaphorcan be
uprooted from their original context only at the risk of thereby being
killed as metaphors. Relative to their present context of interpretation,
these metaphors are dead.
There is also a second feature to the deadness of root-inized metaphors.
As I noted in chapter 3, one essential feature of indexicals and demon-
stratives is that their interpretation, that is, their content, is always xed
by their character relative to their actual context of utterance. There are
no operators on character, operators that shift their context of interpre-
tationthe context in which their content is generatedfrom their con-
text of actual utterance to a counterfactual context. In chapter 6, I argued
that a similar principle holds for metaphors. Their interpretation always
clings to the context set of presuppositions in their actual context of
utterance. To the extent to which root-inized metaphors violate this
principle, they are also, then, not quite metaphors. Their character
is more constant than that of live metaphors. Instead the role of the
presuppositions in which their content is ``rooted'' is more like the pre-
semantic status of the presuppositions that determine the contents of
proper names. Dead metaphors of this root-inized type are on the way to
becoming name-like.
In sum, routinized metaphors have lost their aliveness because, with the
same presuppositions in each of their contexts of utterance, they have a de
facto constant content in all contexts. Root-inized metaphors have lost
their aliveness because, rooted in one counterfactual context, they have
314 Chapter 8

lost the potential for alternative interpretations in alternative (including

the actual) contexts of utterance; their character has lost, you might say,
its soul. The third, and last, class of metaphors I wish to discuss are not
dead metaphors in either of these two ways, and it is also not clear that
they correspond to any of our pretheoretical ideas of dead metaphors.
However, they count as dead, or at least as not fully alive, on my crite-
rionand for a reason that is common to many metaphors in some
form or to some degree. I'll call them acquired as opposed to introduced
metaphors.
3. Acquired metaphors. Suppose that a speaker, Jack, hears another, Jill,
say that Max is a Turing machine (no insult intended). Jill presupposes
that Turing machines are very basic, rather unexciting rule-governed
computational devices, some of whose operations are even hard-wired in,
and she intends her use of the metaphor to express properties like being
unexciting but reliable, steady and even predictable, solid, methodical,
disciplined to act according to step-by-step rules to the point of boredom.
Jack, on the other hand, knows that Turing machines are computers but
nothing more than that. But Jack does know, and knows that Jill knows
and knows that she knows that he knows, that Max is extremely orderly
and methodical, that he is great at solving problems step-by-step, that he
never misses a step, that he is not especially interesting as a personality, in
fact that he is a real bore, but that he is utterly reliable, someone you can
count on. Jack also believes that Jill is using the predicate `is a Turing
machine' metaphorically, and he assumes that Jill is familiar (more than
he is) with Turing machines and intends to express certain properties true
of Max by her utterance in accordance with her presuppositions about
Turing machines and about Max. Now, although Jack does not himself
share Jill's presuppositions about Turing machines, he observes Jill's use
of the metaphor (possibly applied over time to others besides Max, this
being one of Jill's favorite metaphors) and gradually becomes able to
apply the metaphor himself. One day, let's suppose, he even exclaims in
exasperation: ``Max is a Turing machine.'' Let's say that in these circum-
stances Jack has acquired the metaphor `is a Turing machine'.
We can characterize an acquired metaphor in these terms. Jack does
not himself possess m-associated presuppositions for the expression `is
a Turing machine'; strictly speaking, then, the predicate has no meta-
phorical interpretation with respect to his individual context set of pre-
suppositions cj ; that is, it does not express in cj a set of properties, or a
complex property, that would determine, for any individual i, that i is or
is not a Turing machine, where the predicate is interpreted metaphori-
From the Metaphorical to the Literal 315

cally. Jack himself could not introduce the metaphor. However, Jack is
able to apply the expression `is a Turing machine' (correctly, most of the
time) as it would be applied if it were interpreted metaphorically. Indeed a
third person, observing Jack's use of the expression, might project onto
him a grasp of its metaphorical interpretation (much as adults sometimes
project onto children their own grasp of a certain application of an ex-
pression as a metaphor). This and more: If Jack acquired `is a Turing
machine' from Jill with the intention of using it with the meaning (char-
acter) she (or the person from whom Jill learned the expression) gave it
that is, with its metaphorical characterand he believes that she was
using it metaphorically, then not only is Jack able to use `is a Turing
machine' as it would be used metaphorically; one might argue that he
has acquired it as a metaphor. Because he intends to use the predicate in
accordance with the intentions of whoever introduced it, and he assumes
it was introduced as a metaphor, his own utterancesdespite the fact that
they lack metaphorical interpretations in his own contextare meta-
phorical or, more precisely, metaphorically acquired.
Acquired metaphors are meant to illustrate the fact that some of our
uses of metaphor are socially determined in ways that have not been suf-
ciently well appreciated in the literature, that we often piggy-back on
others' presuppositions when we knowingly use metaphors. The way in
which Jack's acquisition of the metaphor depends on Jill's presupposi-
tions is reminiscent of Hilary Putnam's (1975) idea of the linguistic divi-
sion of labor: that many ordinary speakers' uses of natural kind terms
depend on or are parasitic on an expert's knowledge of full satisfaction-
conditions for the kind-term. We use such terms with our own incomplete
knowledge by deferring to experts. It is not clear that there are ``experts''
on metaphors, but the same principle of deference holds with respect to
introducers and acquirers. And the more the interpretation depends on
socially accessible contextual presuppositions, the more presemantic
becomes the role of context and the less metaphorically alive.
The three kinds of dead, or dead-like, metaphors I have discussed
routinized, root-inized, and acquiredare examples of three dimensions
of the literal: context-invariance of content, the presemantic status of its
context, and its social character. None of these conditions is itself a su-
cient condition for being literal, nor do I want to claim that they are
jointly sucient or individually necessary. But each condition corre-
sponds to a criterion, or perhaps symptom, we associate with the literal in
our pretheoretical conception.
316 Chapter 8

The dierence between being metaphorical and being literal is not, or

not simply, a matter of standing at opposite poles on a linear continuum
in which metaphors of dierent degrees of vitality and moribundity bridge
the two extremes.20 For one thing, it is never made clear by proponents of
this view what the continuum measures or that what it measures can be
quantied as the picture assumes. Furthermore, the continuum picture
seems to rest on a confusion between the recognition and interpretation
questions. Instead of addressing the question of what makes a metaphor-
ical interpretation dead or alive, the picture addresses the question: How
do we identify whether a given token of a metaphor is dead or alive? We
don't (yet) have a good answer to the recognition question. But it ought
to be clear that we do not perform the recognition task either by feel (e.g.,
degree of Humean vivacity) or by ``looking'' at the concrete surface
expression or at the product of the interpretation without assessing its
structure. Nor do we introspect the process we happened to employ to
arrive at the interpretation. In any case, it is not necessary in order to say
what makes an interpretation metaphorically dead that we explain how
we identify one, nor should we look to an account of its identication for
an explanation of the character of its interpretation.
But if the continuum picture is not right, what is the relation between
the metaphorical and literal? Some think of literal meaning as the afterlife
of a dead metaphor. But that cannot be the whole story: No one ever
claimed that literal meaning only comes from dead metaphorical inter-
pretations. I proposed earlier that as a metaphorical interpretation ceases
to be dependent on presuppositions specic to its actual context of utter-
ance, the metaphorical interpretation dies, and as the interpretation ceases
to be context-dependent at all, it becomes literal. A full depiction of the
literal would be another book. But with a glance back at our discussion of
de re metaphors in chapter 5, let me sketch a dierent picture.

IV Literal Interpretations as Context-independent Interpretations

Metaphors move in families. I have emphasized the degree to which the

interpretation, that is, content, of a metaphor in a context is sensitive to
the networks to which its vehicle is presupposed to belong (in that con-
text). Its exemplication-schema determines what the vehicle of the met-
aphor is presupposed to exemplify and its inductive network is essential to
determining the implications by means of which we typically explicate the
metaphorand manifest our understanding of itby metaphorically
elaborating and extending it. In all these ways the content of a metaphor
From the Metaphorical to the Literal 317

in a context is highly dependent on and sensitive to the other elements in

the various complexes in which it gures. I have also argued that the
exemplication-network to which a given metaphor (vehicle) belongs fur-
nishes it with a cognitive perspective on its content, by which it conveys
information or signicance specic to its character in addition to its con-
tent. Similarly, I have tried to explain the pictorial dimension of a meta-
phor by the fact that its relevant network is sensitive to the slightest
dierence in features of the vehicle. Even the tightest synonymy relation
does not preserve its network associations. Like a picture, no feature of
the vehicle can be discounted as irrelevant to the individuation of its
metaphorical type. For these kinds of considerations, I suggested, in
chapter 5, section III, that on occasion we individuate a metaphor, not
only in terms of its character and content (in a context), but also by ref-
erence to its contextual schema. On that criterion, two metaphors are the
same if and only if their content (in the context), character, and context
set of presuppositions (including ones about schema) are identical. Thus
the network to which a given metaphor belongs and indeed the context in
which it occurs are essential to its identity.
A literal (or one notion of a literal) interpretation is an interpretation, I
now want to propose, that can be assigned to a word relatively inde-
pendently of its context and, in particular, independently of networks to
which the word belongs. When a metaphorical interpretation of an ex-
pression becomes a literal meaning of the word, it becomes the meaning
of the wordthat is, an interpretation that can be assigned to the word
independently of its contextually presupposed networks. Its content
becomes a feature ofpossessed bythe word (type)regardless of its
containing schemas. This is the sense in which that content is literalof
the letter: of the word in isolation from its containing schema or context.
Literal meaning is atomistic, unlike metaphorical interpretation, which, if
not holistic, always depends on its containing context. Hence, once it
becomes literal meaning, a content is assignable by type, the same con-
tent for all replicas of the type (except where the type is indexical or
demonstrative).
In rejecting the continuum picture, I have implied that the literal is not
simply a dead metaphor of the highest degree of moribundity (an inter-
pretation than which there is none that is deader). From the point of view
of my theory, a literal expression is always formally dierent in kind, not
just degree, from a metaphorical expression: Even if the content of
`Mthat[F]' in c is identical with the content of a literal expression C in c,
`Mthat[F]' and C are obviously of dierent types.21 Unlike routinized
318 Chapter 8

metaphors that have a context-invariant content, that is, the same content
in all contexts, a literal expression has the same content regardless of the
context in which it occurs, independently ofapart fromany context.
We don't have to work through the contextual presuppositions associated
with the expression to determine its content on that occasion. When a
metaphorical interpretation of an expression F becomes the literal mean-
ing of an expression C (even where F and C are homonyms), the content
is ``liberated'' from context. Or, more precisely, the content is liberated
from the semantic context. It remains, or perhaps becomes, presemanti-
cally context-dependent.
Parallel to this contextual liberation movement, there is an epistemo-
logical transformation that transpires as the metaphorical interpre-
tation becomes a literal meaning. I argued in chapter 5 that the
context-dependence of, say, exemplication-metaphors enables them to
express contents, or properties, that the speaker-interpreter cannot express
literally, for which he does not have the requisite conceptual resources at
the time to express in a context-transcendent manner. The deciency is
epistemological: What the interpreter lacks is not just a word but the kind
of understanding that is necessary for fully conceptualized, or de dicto,
interpretation. A literal interpretation of an expression, in contrast, is a
fully conceptualized (de dicto) interpretation: If such an interpretation is a
propositional content, then a literal interpretation of a sentence is a con-
tent that contains no bare individuals, no bare properties, but exclusively
their conceptual representations. Thus, as a speaker better understands a
property or set of properties expressed by a metaphor by exploiting its
contextual relations, as the metaphor is integrated into the interpreter's
conceptual repertoire, she acquires the ability to express it apart from
particular objects that exemplify it. The interpreter knows under what
conditions it does and does not apply, apart from the conditions of a
particular context in which it is the property exemplied by a particular
thing.
Talk of the literal as purely conceptualized content may seem at odds
with our talk of literal meaning, meaning rather than content. But a literal
meaning is a meaning that in any context determines the same purely
conceptualized content for an expression. Such a meaning is not merely
context-invariant, or eternal, expressing the same content in every con-
text. It is context-transcendent; the word has its literal meaning not simply
in every context but out of context.22 Of course, this is not to say that the
metaphorical and the literal refer to dierent kinds of content or that they
constitute dierent kinds of thinking.23 The epistemological dierence
From the Metaphorical to the Literal 319

between the metaphorical and literal is not a dierence in what we think

but in how we represent to ourselves the contents of our thought.
At dierent points in this book I have oered suggestions as to how to
understand the formula that the metaphorical depends on the literal. One
interpretation I have not pursued is that the relevant dependence requires
that the literal temporally precede the metaphorical. Indeed the opposite
ideawhich on the face of it seems absurdwas once proposed by no
less than Jean-Jacques Rousseau. In his essay On the Origin of Language,
Rousseau writes that the ``First Language had to be gurative.''24 Per-
haps we can now make sense of his thesis in the terms of my proposed
account. Suppose that the context-oriented character of a metaphor en-
ables us to express and represent to ourselves contents that are less than
fully understood and fully conceptualized, and suppose that complete
understandingthe fully conceptualized kind that is expressed literally
follows incomplete knowledge. In that case Rousseau might have been
right after all. ``Figurative language was the rst to be born. Proper
meaning was discovered last. One calls things by their true name only
when one sees them in their true form.''25
Notes

Preface
1. Johnson (1981), ix; cf. Lako and Johnson (1980), ixx.
2. Compare the exchange between Hesse (1987) and Rorty (1987) on Davidson.
3. Still other authors such as Kittay (1987) supplement classical semantics with
other purported semantic theories, such as semantic eld theory; see ch. 6, sec. IX
on the semantic status of such theories. In any case, these attempts shift the brunt
of their explanation of metaphor away from classical semantics.

Chapter 1
1. The choice of example is not intended to raise special questions connected to
the interpretation of metaphors in ction; for related discussion, see below, ch. 6,
sec. III.
2. The propositional information of a subsentential expression F is the content, or
factor, F contributes to the information of any sentence to which it belongs.
3. I shall sometimes use this awkward phrase speaker-hearer and sometimes just
one or the other of the hyphenated terms; unless explicitly noted, I do not intend
to be exploiting a particular ``perspective.''
4. The classic exposition of this view is Beardsley (1962), although the position
was very widely shared. For references and critical discussion, see Stern (1983).
Some advocates of the deviance condition also tried to use it to explain the inter-
pretation of a metaphor, e.g., by way of transfer, selective weighting, and elimi-
nation of the deviance-producing lexical features contained in the entries for the
constituent words. See, e.g., Beardsley (1978), Levin (1977), Matthews (1971). The
features responsible both for the recognition and interpretation of the metaphor
would, then, be sentence-internal. For critical discussion of this model of inter-
pretation, see below, ch. 6, sec. IX.
5. The phrase ``twice-true'' was coined by Cohen (1975), (1976). Other critics of
the grammatical deviance condition at that time included Reddy (1969) and
Binkley (1976). Diehard defenders of various versions of the deviance condition
include Beardsley (1978) and Kittay (1987); for further discussion, see Stern (1983)
and below, ch. 6, sec. IX.
322 Notes to pages 411

6. Thanks to Mohammed-Ali Khalidi for the reference.

7. One might object to this example that it is so obviously true (taken literally) as
to render it deviant on that ground. This may be true, but unless one assumes a
strong analytic-synthetic distinction, it is not clear why this should be theoretically
more problematic for this kind of example than for the earlier examples that are
also true.
8. It is worth noting that in the psychoanalytic literature on metaphor, we are
often warned that utterances by, say, schizophrenics, which we would take to be
metaphorical in other contexts (if, for example, they were found in poetry), should
not be interpreted metaphorically but literally.
9. See Recanati (1995), Glucksberg and Keysar (1993), Gibbs (1984), (1989), and
Rumelhart (1993). Although many psychologists have challenged the serial literal-
rst model on the grounds that the processing time for metaphorical sentences is
no longer than that for literal sentences, Recanati (1995) and Winner (1988)
challenge this evidence, taking into account the possibility of masking eects.
However, for independent reasons they favor a parallel-processing model, and in
ch. 4, sec. IV I shall also sketch a model of metaphorical interpretation that sup-
ports this approach to metaphorical recognition.
10. Culler (1981), 1920. Culler cites the example to illustrate the problem of in-
terpretation, although he does not distinguish recognition from interpretation. He
also proposes that the example supports a Davidsonian approach, yet notes that it
presupposes a structure of interpretation that would be dicult to incorporate into
Davidson's framework.
11. Nonetheless much of the literature on metaphor persists in conating the rec-
ognition and interpretation questions; see, e.g., Black (1962), 25; Black (1993),
3335; Beardsley (1958), 161; Ortony (1993b), 5; Sadock (1993), 478; and Grice
(1975), 71. In ch. 2 I shall classify this role of the context as presemantic insofar as
it enters into assignment of a metaphorical type to an utterance.
12. There are many steps in this argument at which one might balk; see below, ch.
7. I mention the argument here only to motivate the structure of the book.
13. I use feature, which is intended to be a neutral term, rather than, say, prop-
erty, in order not to beg the question whether what is conveyed by a metaphor is
proposition-like or a propositional constituent.
14. Among the dissenting minority are Matthews (1971), Cohen and Margalit
(1972), L. J. Cohen (1993), Kittay (1987), Bergmann (1979), (1982), Berg (1988),
and Leezenberg (1995). On the rst four theories and how they dier from my
account, see below, ch. 6, sec. IX. The last three views are those most similar to
mine in the literature.
15. Rorty (1987), 285286. Here Rorty claims to be oering an explication of
Davidson's position; on dierences between their views, see ch. 2, sec. III. For a
similar conception of the role of context in metaphorical interpretation, see
Scheer (1979); Elgin and Scheer (1987); and Stern (1988).
16. Davidson (1984), 245.
17. See Margalit and Goldblum (1994).
Notes to pages 1122 323

18. According to a variant text, the saying reads ``Before [the priest] Eli's sun had
set, the sun of [the prophet] Samuel had risen'' (BT Qedushin 72b); for our pur-
poses, the interpretation is more or less the same.
19. Examples (5) and (6) (slightly modied) originated with Avishai Margalit.
20. Carey (1981), 11. I am indebted to Arthur Danto for bringing these examples
to my attention.
21. Cf. Sadock (1993), 44.
22. It is in terms of this close conceptual and formal parallel I wish to draw be-
tween metaphors and demonstratives that my account most diers from other
contextualist semantic theories like Bergmann (1979), (1982).
23. Here I use non-/constant character where Kaplan uses non-/stable character to
express the identical notion. I have departed from Kaplan's own terminology to
avoid a potential misunderstanding that might result from the term ``nonstable''
that could be taken to mean that the character assigned to the expression itself
changes from context to context. What is nonstable about a character for Kaplan
(and nonconstant for me) is the fact that the content it determines can change or
dier from context to context.
24. Cf. White (1996), 88.
25. On the other hand, in his original (1962) paper Black sometimes takes the
metaphor to be the constituent word and in yet other passages talks as if it is not
the individual word but ``the system of associated commonplaces'' that is meta-
phorical.
26. Original credit for the fundamental insight that it is always a whole schema or
family of expressions that is interpreted (or transferred) metaphorically, never an
individual expression, should, however, be given to Goodman (1976); on this
theme, see below, chs. 5 and 7.
27. Black may also be concerned to capture the idea that, although the other
constituents in the sentence (the so-called frame) are not metaphorical in ``the
same way as the metaphorical expression'' (or focus) that undergoes, say, a
change of extension, nonetheless they also undergo some change of interpretation.
Hence in some sense the whole sentence is metaphorical. In reply I would argue:
(1) insofar as we are concerned with propositional content, any changes, however
signicant, that are only emotive or attitudinal lie outside the scope of our
story; and (2) if indeed there are changes in the propositional interpretation of
other expressions in the sentence, there is no reason not to take each such expres-
sion to be metaphorical (in which case the sentence will simply contain multiple
metaphors).
28. White (1996), 202. All the cases I describe in the previous three sentences in
the text are raised by White as problems for a word-focused theory of metaphor
and, in particular, for theories like Black's that (White claims) are limited to tak-
ing a metaphor to be a unique, simple (one-word) focus in a sentence. Although
White may be right that Black's focus/frame apparatus is inadequate to describe,
let alone explain, these more complex phenomena of metaphor, I shall try to show
in chs. 5 and 6 how my semantic theory can capture them. I would add that
324 Notes to pages 2330

White's presentation of Black is not as black and white as he suggests: If one takes
into account Black's own emphasis on the system (of associated commonplaces)
relative character of metaphorical interpretation, his theory need not be inter-
preted as exclusively one-word-based as White alleges.
29. To anticipate a potential misunderstanding: We can either treat ``metaphori-
cal expressions'' such as `Mthat[F]' in the technical sense of ch. 4 as lexically
complex expressions or treat the metaphorical interpretation of the expression F
as the literal interpretation of the metaphorical expression `Mthat[F]'.
30. Kronfeld (1980/81); cf. also Sweetser (1992).
31. See, e.g., Beardsley (1976).
32. See Kronfeld (1980/81) and, now, White (1996) for a sustained philosophical
critique of this kind.
33. See Lako and Johnson (1980), Lako and Turner (1989), and Lako (1993).
34. For complementary thoughts about the evidence for linguistic theories in
general, see Hornstein (1984), 1012.
35. Plimpton (1976), 120121.
36. Margalit (1978).
37. I owe these examples to Lako and Johnson (1980), 4.
38. It is especially risky to speculate on the basis of surface ``appearance'' whether
a given expression is a dead metaphor. Kronfeld cites Alston's use of `fork in the
road' as a nice example of fallacious armchair theorizing. Although Alston tells a
prima facie plausible story of how the phrase came to be metaphorical, in actual
fact it did not historically originate as a metaphor but rather as a literal applica-
tion of `fork' meaning `that which branches or divides'. Likewise, Brooks (1965a)
cites, as an example of a dead metaphor brought back to life, the following
line (quoted by Dorothy Sayers) from a nineteenth-century Oxford poem on the
Israelites crossing the River Jordan dry-shod. When the bearers of the Ark
stepped into the river, the waters suddenly rolled back, ``And left the astounded
river's bottom bare.'' This is a good live metaphor, but the expression `bottom
of the river', as it is generally used, is not a dead metaphor but a straightforward
literal use. Hence this is also not an example of a resuscitated dead metaphor.
39. Cf. the entry for ``Metaphor'' in Preminger (1965), 136141. However, an
exception to this rule are White's (1996) arguments, building on his rich knowl-
edge of poetry and literature, against philosophers' simplistic, single-word exam-
ples of metaphors. Yet, as I indicated earlier, I think many of his philosophical
objections can be answered.
40. Aristotle does not reveal what he means by ``genius'' but one gloss might
be the Kantian conception of genius, as the capacity to produce things that
are inexplicable by rules, yet make sense. Cf. Cohen (1975), 671, who explains
that metaphors are products of genius meaning that they are ``not accomplished
in terms of statable rules''rules of the kind presumed to underlie linguistic
competence.
41. Such a view is held by Isenberg (1963).
Notes to pages 3036 325

42. In denying these distinctions, I do not, of course, mean to deny that there re-
main signicant dierences between the metaphors of poetry and of ordinary
speech. But these dierences are not a function of dierent underlying com-
petencies. Instead they are a function of dierent uses of a common competence to
create dierent eects and products, a dierence like that between the literal lan-
guage used to write a shopping list and that used to write the Gettysburg Address
or Critique of Pure Reason. I would argue that as competent speakers, we all have
a mastery of metaphor, but that only some of us are masters of metaphor.
43. Langer (1942), 112.
44. Otto Neurath (commenting, incidentally, on the early Wittgenstein), cited
by Carnap (1963), 29. This view, which begins with metaphor and eventually
encompasses all language, probably originates with various medieval theological
conceptions of language; for one prehistory, see Stern (2000). On the naturalistic
study of language and its humanistic critique, see also Higginbotham (1982), 156
157.
45. To the extent to which they require explicit learning, peripheral metaphors are
similar to dead metaphors whose interpretations are not grasped via general rules
of context-oriented metaphorical interpretation (rules that require no learning)
and must also be explicitly learned. However, interpretations of dead metaphors
are learned one by one, whereas it is additional rules of interpretation that are
learned in the case of peripheral metaphors. Thus core living metaphors are sur-
rounded on the one side by dead metaphors and on the other by living but
peripheral metaphors. Both dier from the core with respect to the kind of learn-
ing involved in their interpretation.
46. For examples and discussion of some of these additional skills and techniques,
see Hrushovski (1984).

Chapter 2
1. Davidson (1978/1984), 247. All references to this paper will be to the 1984 re-
print and will be cited in the text as WMM.
2. Other members of Davidson's camp include Cooper (1986); Blackburn (1984),
171179; and Rorty (1987).
3. To anticipate a potential misunderstanding: In light of my own account in
ch. 4, it will turn out that Davidson's claim (1a) is compatible with my own view,
that ``metaphorical expressions'' of the form `Mthat[F]' composed out of individ-
ual (simple) words F have metaphorical meanings. The reason is that although
metaphorical expressions lexically represent metaphorical interpretations of indi-
vidual simple words, they are themselves complex expressions.
4. Davidson (1986), 433446. All in-text references to this paper will be cited as
NDE.
5. Davidson (1984), 279; see also ibid., xix.
6. Searle (1993); Grice (1975).
7. Grice (1975), 71.
326 Notes to pages 3743

8. Grice assumes a linear literal-rst model of interpretation (of the type I ques-
tioned in ch.1), and, more important, he also seems to suppose that we always at
least attempt to interpret an utterance that cannot be taken literally rst as an
irony and, only when that fails, as a metaphor. At the same time, he allows for
combination metaphorical/ironic interpretations, in which the content of the
utterance interpreted metaphorically is also meant ironically.
9. A similar criticism applies to Martinich (1984) whose use of salience fares no
better as a candidate for meaning than Grice's use of resemblance. Likewise, his
introduction of the maxim ``Be relevant,'' though relevant, is hardly sucient to
do all the work he wants it to do.
10. Compare Davidson's objection to positing a gurative meaning for similes:
``The point of the concept of linguistic meaning is to explain what can be done
with words. But the supposed gurative meaning of a simile explains nothing; it is
not a feature of the word that the word has prior to and independent of the con-
text of use, and it rests upon no linguistic customs except those that govern ordi-
nary meaning'' (WMM 255).
11. For a more detailed discussion of these Davidsonian themes, see Stern (1991).
12. Davidson (1984), 273274.
13. On the connection between communication, literal meaning, and truth-
conditions, see Davidson (1984), 45. Note that Davidson takes a single type of
semantic interpretation, e.g., truth-conditions, to correspond to the required kind
of understanding, or interpretation, thus leaving no allowance for partial degrees
of understanding. This may have the advantage of simplicity but it is also a poten-
tial source of problems.
14. More specically, I 's theory is of what he expects S intends to say with his
(S's) particular words on this occasion, given I 's prior knowledge of S and of the
rst meanings S has previously attached to his (S's) words. S's theory, on the
other hand, consists of his intentions that particular words of his will be inter-
preted by I as saying such-and-such, given his beliefs and his expectations about
I 's ability to interpret him (S) as saying those things with those words. So, if S
believes that I will not be able to interpret him as saying such-and-such with cer-
tain words, he will not intend for those words to be interpreted in that way.
15. On this provocative claim, and some philosophers' reactions to it, see Hacking
(1986) and Stern (1991).
16. Here I gloss over a number of Davidson's unsupported, and potentially
problematic, claims about the means-ends ordering of intentions.
17. Despite its initial characterization as a ``preliminary stab,'' there is nothing
preliminary about Davidson's continued use of the notion of rst meaning
throughout NDE as the explication of literal meaning; see, e.g., NDE 442.
18. There may be exceptions to this rule. Knowledge of the secondary intention of
an utterance (e.g., knowing that it is a promise or threat) may aect which rst
meaning is assigned to a constituent word; in this way, assignment of rst meaning
may also be sensitive to postsemantic contextual features that bear on the utter-
ance's secondary intention.
Notes to pages 4344 327

19. Cf. Davidson's principle of the autonomy of linguistic meaning, in Davidson

(1984), 164; cf. also 274. These notions of autonomy of meaning should, however,
be distinguished from the notion I discuss in section V.
20. Davidson's argument should be distinguished from another argument in the
literature, which treats metaphor as a distinctive kind of force or illocutionary act
that runs on a line parallel to but dierent from assertion, questioning, lying,
promising, criticizing, and so on. Just as assertion is a use of language in which
one represents oneself as intending to say what is true; lying, a use representing
oneself as believing what one does not; and promising, a use with the purpose of
representing oneself as undertaking an obligation to perform some act; so meta-
phor would be a use of language in which one represents oneself as calling atten-
tion to certain similarities or other features. This is not simply to say that (many)
metaphors in fact do direct our attention to similarities; that much is indisputable.
The claim is that metaphors constitutively perform a distinctive illucutionary act,
an act these writers roughly describe as ``calling our attention to a certain like-
ness,'' or ``inviting'' us to ``appreciate'' a resemblance, or ``inspiring'' a certain
vision, or ``proposing'' that things be viewed a certain way. These ``proposals'' or
``invitations'' or ``attention-calling'' acts may be apt or distorting, heuristically
valuable or misleading, suggestive or confusing, butand this is the crux of the
claimthey are neither true nor false. Hence, if assertions are either true or false,
the dening function of a metaphor excludes assertion. On these non-assertional
functional characterizations of metaphor, see Loewenberg (1975), 174; Black
(repr. 1981), 75 (who describes metaphor as ``organizing our view'' of the subject);
and even more explicitly, Black (1993), 4041. On the other hand, Fogelin (1988)
allows for both information-giving and attention-calling uses for metaphor.
Undoubtedly some metaphorsin poetry and scientic contextsare uttered
with and only with such a non-assertional heuristic force. But metaphors occur in
utterances whose sentences express all the grammatical moods, e.g.,
The Declarative: Juliet is the sun.
The Interrogative: Is Juliet really the sun?
The Imperative: For God's sake, Juliet, act (like) the sun!
The Optative: Would the sun kindly enter.
Indeed, we can use a metaphor to perform a speech act or utterance with any
forceeither the direct force corresponding to the mood of the sentence or some
other indirect forcein addition to the act of drawing our attention to the resem-
blance (or other feature) that enters into the content of the metaphor. The fact
that metaphors call likenesses to our attention does not exclude the possibility that
they also make assertions.
Furthermore, one might argue that in order to explain why metaphors call
(only) specic resemblances, analogies, or features to our attention, and not
others, we must appeal to assertion (or some other act presupposing assertion).
The idea would be that we assert propositions not simply that we believe are true
but also that we believe are not already common knowledge or shared pre-
suppositions in the context. Hence we survey the background or context for the
kind of property or resemblance that would make Romeo's assertion informative
as well as true. This additional requirement for assertion serves, then, to eliminate
328 Notes to pages 4550

or lter out those resemblances or aspects that fail to serve the additional asser-
toric purpose of the utterance, thus explaining why only some and not other
resemblances or features are expressed by a metaphor.
Finally, one might object to this argument that, for this purpose, truth-values or
truth-conditions are not necessary as the constraints on metaphorical interpreta-
tion; weaker conditions like warranted assertability or acceptability, or even
something weaker like ``making sense,'' would also do the job. I am sympathetic
to this objection, but it applies to one's theory of meaning as a whole. If we were
to turn our general theory into a theory of, say, warranted assertability, analogous
revisions could (and would) be made for metaphor.
21. This, despite Davidson's tendency to conate, or interchange, ``literal mean-
ing'' and ``literal truth-conditions''; see, e.g., WMM, 247.
22. On the comparison to jokes (and riddles), see Aristotle (1984), Rhetoric
1412a 181412b 32; Cohen (1978), (1983); and below, ch. 7, sec. V.
23. Davidson's use of multiple expressions in WMM to refer to the relation be-
tween the utterance of the metaphor and its ``eect''e.g., ``makes us see,''
``alerts,'' ``inspires,'' ``leads us to notice,'' ``prompts,'' ``draws our attention to,''
``provokes''may make one wonder whether there is a single relation at work,
but in any case his ``causal'' account is epistemic. It should therefore be dis-
tinguished from Max Black's notorious claim that metaphors ``create'' similarities
(1962, 37) by which he means, as he insists (in Black 1993, 3538), not only reveal
but also constitute or bring into existence. For yet another sense of how a meta-
phor may ``make'' (us recognize or notice) a similarity, see below, ch. 5, sec. II.
24. To support his reply, Davidson would, to be sure, appeal to Quine's
rejection of the analytic-synthetic distinction. But the persistent objector might, in
turn, reply that what is unacceptable about the analytic-synthetic distinction is its
epistemological use to ground an a priori/a posteriori distinction, which is not a
burden the distinction is made to shoulder here. For dierent and extended dis-
cussion of replies to the objection, see Margalit and Goldblum (1994), 235237,
and White (1996), 204226.
25. For this reason, White's claim that ``to apprehend a metaphor as a metaphor
involves ignoring whatever literal sense [the sentence] may have'' (1996, 226) may
be too strong if that involves also ignoring its syntax; in any case, I have not been
able to identify a specic argument for this particular conclusion in White's subtle
examination of examples.
26. I am not raising the general question of the relation between truth-conditions
and meaning, either to criticize or defend Davidson's program. For discussion, see
Davidson (1984), especially essays 2, 4, 9, and 12, and Davidson's postscripts and
references therein.
27. See Danto (1981), 179189, who concludes from the argument (of the previ-
ous paragraph) establishing the non-extensionality of metaphorical contexts that
they are indeed intensional contexts. For an extensionalistic explanation of the
work done by intensions or meanings to account for the failure of substitutivity,
see Elgin and Scheer (1987) and, for critical discussion, Stern (1988). For further
discussion of Goodman's theory of metaphor, see also Stern (1997).
Notes to pages 5157 329

28. Time, Feb. 28, 1985. On cross-cultural dierences of metaphorical interpreta-

tion, see Asch (1961).
29. On the role of context more generally in metaphorical interpretation, see
Scheer (1979).
30. Rorty (1987), 284; cf. also 296. For the position Rorty is contrasting with
Davidson's, see, e.g., Hesse (1987). Davidson emphasizes that his noncognitivist
view does not mean (as it did for many earlier positivists) that metaphor is ``con-
fusing, merely emotive, unsuited to serious, scientic, or philosophic discourse''
(WMM, 246), but his praise of metaphor never reaches the exalted heights of
Rorty's; at best Davidson thinks ``metaphor is a legitimate device not only in lit-
erature but in science, philosophy, and the law'' (ibid., my italics).
31. See Davidson (1967/1980), 149162.
32. However, see Davidson's most recent statements on this issue: ``It is clear that
there may be no clue to the character of an appropriate law in the concepts used
on some occasion to characterize an event. What may be the case is that if a sin-
gular causal statement is to be explanatory in some desired sense, it must put its
hearer in mind of at least the general nature of a relevant law'' (1995, 265).
33. A number of brief comments: (1) In the quoted passage Davidson makes the
same claim for irony as for metaphor, and indeed I think his analogy works better
for irony. However, in ch. 6 I'll argue that the two gures ought to be explained
dierently; so I shall simply ignore irony here. (2) Both Davidson's exposition of
Donnellan's idea of a referential denite description and his characterization of
malapropism as a change of meaning are not beyond challenge; e.g., his focus on
referential descriptions that do not semantically designate the intended referent is
misleading, albeit admittedly inspired by some of Donnellan's original examples.
Nonetheless for the sake of argument I shall grant him his way of presenting these
two subjects. (3) As Roger White has pointed out (pers. comm.), Davidson's
remark in the quotation that the speaker says ``something true'' (with the false
literal sentence) apparently conicts with his well-known thesis that metaphors do
not convey ``a denite cognitive content'' or propositional content and, therefore,
ought not to be either true or false. What to make of this slip is not clear. (4) In the
quotation Davidson does not elaborate on the analogy between metaphor and
referential descriptions. It might be argued in his defense that he did not mean
anything stronger than to note a similarity in passing and, in particular, that he
did not mean to propose the referential denite description as a model by which to
understand metaphorical use and its dependence on the literal. In reply I would
rejoin that throughout NDE Davidson emphasizes the distinction between literal/
rst meaning and speaker's meaning, illustrating each (with malapropisms, refer-
ential descriptions, etc.) and leading his reader to think that he intends his claims
to be generalizable. It is in that spirit that I suggest this second Davidsonian model
of metaphorical/literal dependence.
34. I owe the example to Sidney Morgenbesser. One might object that the exam-
ple is not a malapropism but a semantic error; on Davidson's characterization,
however, it counts as a malapropism. On distinguishing malapropism from other
kinds of semantic speech errors, see Fay and Cutler (1977).
330 Notes to pages 5867

35. I am indebted throughout this paragraph to discussion with Lauren Tilling-

hast.
36. Here I assume, as I'll argue in ch. 3, that demonstratives are directly referen-
tial terms; i.e., their propositional content (in a context) or truth-conditional factor
is the individual to which they refer rather than a Fregean sense, concept, or
intensional entity.
37. See Davidson's comments in (1984), essays 2, 3, 4, and 9; and Weinstein
(1974). Davidson introduces the relativization to time to account for tense, even
apart from occurrences of the indexical `now'.
38. Because Davidson's own apparently preferred treatment of demonstratives
relativizes the truth-predicate for sentences to contextual parameters, his strategy
makes him vulnerable to some of the same objections that Kaplan raises against
other relativized-truth accounts, like the index-theories of R. Montague and
D. Lewis: see Kaplan (1989a), 507510. An alternative treatment of demonstra-
tives within absolute truth-theory, proposed by Weinstein (1974); Burge (1974);
Taylor (1980); Higginbotham (1988), (forthcoming); and Larson and Segal (1995),
takes utterances rather than sentences as the truth-vehicles, but conditions the
T-sentence on an antecedent clause that species the relations that must hold
between the contextual values and sentence in order for the utterance to have
truth-conditions. This strategy avoids Kaplan's criticisms but raises large issues
for our understanding of the notion of truth.
39. For the line of argument in this paragraph I am indebted to Roger White.
40. Note that this notion of autonomy of meaning is entirely dierent from
Davidson's principle of the same name (see note 19 above).
41. A full explanation of how we should technically capture this dependence of
the metaphorical interpretation of an expression F on its literal meaning will,
however, have to wait until ch. 6.
42. See Higginbotham (1989), 159165.
43. Cf. Higginbotham (1989), 162.
44. There is also the question of how conditions like (A) or (I) could even be
learned on accounts like Davidson's. Because the conditions in question primarily
exclude interpretations the sentence does not and never could have, and these
excluded interpretations are presumably never part of the linguistic data to which
speakers are exposed, the conditions cannot be inductively learned from conrm-
ing instances. On the rationalist consequences of these factsthe ``absence of
negative evidence'' as it is known in the psycholinguistics literaturesee Higgin-
botham (1989), 162. and Hornstein (1984), 47.
45. The contrapositive reasoning I have just rehearsed is meant to be reminiscent
of one of Chomsky's (1980) arguments for the so-called competence-performance
distinction and his related argument that a theory of linguistic competence proper
will explicate only what the speaker knows about specically linguistic properties
that contribute to his complex ability that issues in acceptable speech; see also
Chomsky (1968), (1988). For yet another notion of meaning as ``partial'' under-
standing, see Spolsky (1987/88).
Notes to pages 6870 331

46. On the latter, see Hills (1997).

47. To anticipate a possible objection: An empirical distinction between knowl-
edge of, or competence in, language and extra- or nonlinguistic abilities and
knowledge need not entangle us in the epistemological quandaries of the analytic-
synthetic distinction. On this issue, see also Chomsky (1988) and Higginbotham
(1989).
48. Here I follow Williams's (1977) analysis of verb phrase anaphora involving an
interpretive (or copying) rule of Discourse Grammar rather than a deletion rule of
Sentence Grammar; cf. also Chomsky (1968), 3335; and Kempson (1977), 128
132.
49. According to my semantic theory, as I'll argue in ch. 6, the violations in (14)
and (15) are formally dierent. In (14) the antecedent, which is literal, and the
anaphor, which is metaphorical, will be formally assigned dierent characters:
(14a) *The largest blob of gases in the solar system is the sun, and Juliet/Achilles
Mthat[`is the sun'].
In (15), however, the antecedent and anaphor, both metaphors (indeed both
sharing the same metaphorical expression type), will have the same character.
(15a) *Juliet Mthat[`is the sun'], and Achilles Mthat[`is the sun'].
In (14) the violation on copying is at the level of character; in (15) at the level of
content. In fact, there is some evidence that the dierence in ``level'' at which the
copying constraint is violated aects our intuitions of unacceptability. It is possi-
ble to cancel the implication of univocality at the level of content, as in (15), but
not at the level of character, as in (14). (Thanks to Malka Rapaport for these
examples.) Contrast (14b) and (15b):
(14b) *The largest blob of gases in the solar system is the sun, and Juliet/
Achilles is, toobut not in the same sense/way.
(15b*) Juliet is the sun, and Achilles is, toobut not in the same sense/way.
To avoid a possible confusion, note also that
(15c) *Juliet is the sun and Achilles is the sun
whose underlying interpretation is also (15a) will be unacceptable for the same
reason (and more or less to the same degree) as (15). Although both occurrences of
`is the sun' are metaphorical and therefore share the same character, in order for
(15c) to be acceptable (and, say, true), one of the following three alternatives
would have obtain: either (i) both occurrences would have to have the same con-
tent, given the same context (as we shall argue in ch. 4: the same set of contextual
presuppositions); or (ii) each conjunct would have to have a (possibly dierent)
content (under which it is true) but be determined relative to the same (set of )
context(-ual presuppositions); or (iii) each conjunct would have to have an
acceptable (and true) content relative to a possibly dierent set of contextual pre-
suppositions. Alternative (i) cannot be presumed without special pleading, given
our common knowledge of Juliet and Achilles, common knowledge that will be
part of the context. Alternative (iii) would require that the context change during
the utterance of the single sentence, a possibility that also needs special pleading.
Alternative (ii) is ruled out for the same reason either as (i) or (iii). I shall return to
332 Notes to pages 7178

discuss this example in ch. 6, sec. IV, but for now I would emphasize, in order to
motivate the need for metaphorical meaning, that what matters is the impossibility
of literal/metaphorical crossover of interpretation. (Thanks also to Jay Atlas for
discussion of these matters.)
50. It should be observed that on Davidson's account both Romeo's and Paris's
utterances are absurdly false according to what they literally meanwhich is all
they mean according to Davidson. And insofar as they have the same truth-value,
it is not clear how he can even describe Romeo's and Paris's disagreement.
51. Note that my argument is not that metaphorical interpretations are (literally)
innite in number, hence that there is no nitistic base on which to rest composi-
tionality for a language containing metaphors. Although some have taken that
line, there is no evidence that any expression, no matter how many interpretations
or meanings it has, has an innite, or truly unbounded, or even indenitely or
incommensurately large, number of simple interpretations. For discussion, see
Ross (1981), Lycan (1987), and Grandy (1990).
52. I owe these examples to Cynthia Welsh. The two meanings of `ear' seem to be
entirely unrelated historically: The hearing organ ear derives from Indo-European
`ous-', which refers to the bodily part; `ear [of corn]' derives from Indo-European
`ak', which refers to a sharp side or point. The case of `corn' is slightly dierent:
The two meanings of `corn' seem to have roots with similar meanings (`greno',
meaning grain, and `ker', meaning to grow), although they are entirely unrelated
nowadays.
53. Cf. Chomsky (1977), 6769.

Chapter 3
1. See Kaplan (1989a,b), selections of which are now reprinted in various collec-
tions, e.g., Ludlow (1997); see also Kaplan (1973), (1979), and (1990) for ancestors
and descendants of the main monograph. For reasons of space, I have limited this
chapter to the two main themes that bear on my application to metaphor; I take
up additional topics in later chapters as they apply to metaphor. I also use the
notes, more in this chapter than in others, to indicate complications, qual-
ications, and problems I could not address in the streamlined exposition in the
text. Readers unfamiliar with Kaplan's monograph are strongly urged to study the
wonderfully clear original to get a sense of the theory as a whole.
2. Where there is no explicit verbalized linguistic modier for the pure demon-
strative, I would argue that one is tacitly supplied in context. As a consequence, all
complete demonstratives of surface form `That[D]' where D stands for a non-
linguistic demonstration have as their underlying semantic form: `That F [D]'
where `F ' is a placeholder for some predicate.
3. The theory that was criticized was only questionably held by Frege or by any
other historical gure. But one positive by-product of the critique was that it
stimulated a renaissance in Frege studies, among them and especially relevant to
my account, Burge (1977), (1979a). My aim here is not to set the record straight or
to argue for one or another theory, but simply to provide necessary background.
Notes to pages 7986 333

4. See Kripke (1980), Donnellan (1966), Putnam (1975), Kaplan (1989a), and
Perry (1977).
5. On rigid designation, see Kripke (1980) whose analysis of naming in terms of
rigid designation was subsequently criticized by Kaplan (1989a), 492497;
(1989b), 568576; and Almog (1986).
6. See Burge (1977), 354362; Perry (1977); Evans (1981), (1982). A classic alter-
native to perspectivalism would be Russell's (1956) notion of direct acquaintance.
7. For criticism of this strain in the New theory, see Wettstein (1986); Taschek
(1987).
8. Which is not to say that each purely conceptual, qualitative representation will
necessarily be expressible verbally, say, by an eternal complete denite descrip-
tion.
9. For possible diculties with this formalization, see Braun (1995).
10. On my choice of terminology ``nonconstant'' instead of Kaplan's own term
``nonstable'' see ch. 1, n. 23. Note also that in assigning characters to all expres-
sions across the board, we encounter a major diculty for Kaplan's theory, as he
already notes at the end of (1989a), 558563. In the case of proper names, and
indeed all eternal directly referential expressions, ``all three kinds of meaning
referent, content, and charactercollapse''; hence ``proper names do not seem to
t into the whole semantical and epistemological scheme'' based on the character-
content distinction (562). For dierent solutions for this problem, see White (1982)
and Almog (1984).
11. For this terminology, see Almog (1986).
12. In addition to the dierences between direct reference and rigid designation
noted in the text, it has also been argued that rigid designation is primarily a
metaphysical notion and direct reference, semantic; see Kaplan (1989a); Almog
(1986); and, for criticism, Breen (1993).
13. On logical constants, see Salmon (1990); on ordinary proper names, see
Kaplan (1989a), 558563, and Almog (1984).
14. Kaplan (1989a), 494497 originally proposed this ``metaphysical'' notion of a
singular proposition as a ``more vivid'' alternative to the possible worlds concep-
tion, which he nonetheless continued to employ in the formal Logic of Demon-
stratives. For subsequent criticism that renders the metaphysical conception
inevitable, see Salmon (1990); Soames (1988); and Kaplan's (1989b) own reec-
tions, 579, n. 28.
15. On natural kind and substance terms, see Kripke (1980); Putnam (1975). On
the syntax of predicate demonstratives, see Jackendo (1990), 4851, and Han-
kamer and Sag (1976).
16. One might object that the predicate demonstrative should really be analyzed
as `the shape of that' since we demonstrate the given shape by pointing at an
object whose shape it is; here the singular demonstrative would be rigid. But this
fails to give the correct content for `is that shape'. Even if the demonstrative `that'
is rigid, its demonstratum might have dierent shapes in dierent circumstances.
334 Notes to pages 8694

But the propositional constituent of `is that shape' is the actual shape of the dem-
onstratum. So, `is that shape' at best could be analyzed as `is the actual shape of
that' or, more explicitly, as `has the shape S such that it is actually the case that S
is the shape of t' where t is the direct referent of the occurrence of `that' in the
context. But even so, it is the shape that is the content of the phrase, and since
its extension may vary over circumstances, the predicate is nonrigid, hence not
directly referential.
17. I have not, to be sure, said what a property is, but by the same token, I also
have not said what an individual is. For now I defer these metaphysical questions.
18. One might also argue that, on pain of regress, there must be some predicates
that are directly referential, i.e., do not x their extension or reference by way of
satisfaction of a descriptive condition. In the text, however, I focus on the epis-
temological character of direct vs. indirect reference because this will be the per-
tinent issue for metaphor in chs. 5 and 6.
19. Despite the dierence I shall point out between indexicals and dthat-
descriptions, knowledge of the character of the latter is also something its speaker
knows solely in virtue of his knowledge of language. However, this is not the case
for complete demonstratives whose characters are a function of the characters of
their accompanying extralinguistic demonstrations, e.g., visual presentations.
Furthermore, if we think of the appearance that constitutes a presentation as
something like a picture (as does Kaplan 1989a, 526527), then the mode by
which it ``presents'' the demonstratum diers sharply from the mode by which a
linguistic expression refers to its referent. For some of these dierences, see my
Goodman-inspired discussion in ch. 7, sec. IX and in Stern (1997). In any case,
this is one respect in which complete demonstratives dier from both indexicals
and dthat-descriptions.
20. For good surveys of the alternative strategies, see Soames (1986) and Reca-
nati (1996).
21. See Kaplan (1989a), 522523, 541553; Kaplan (1989b), 591598; and the
exchange between Crimmins (1992b) and Richard (1993a).
22. I gloss over the fact that only `I' and `here' are treated in Kaplan's Logic of
Demonstratives as directly referential constants, while `now' and `actually' are
treated as sentential operators; cf. Kaplan (1989b), 580, n. 30.
23. Optional parameters can be added beyond these four that constitute a mini-
mal (semantic) context, e.g., an addressee for `you' or temporal indexicals such as
`today' and `yesterday'.
24. For discussion, see Stern (ms.).
25. For reasons of space I cannot address the question for complete demonstra-
tives in this book, although what I say about dthat-descriptions has denite
implications for nonlinguistic demonstrations. In his (1989b), 579582, Kaplan
distinguishes two possible relations of the pure demonstrative `that' to its ``com-
pleting'' demonstration. (1) `That' is a functor or operator for which the demon-
stration is an argument or operand; the demonstration syntactically completes, or
saturates, the otherwise empty argument-place in the demonstrative functor.
Notes to pages 9498 335

(Compare, however, Kaplan's second thoughts about his ``unfortunate'' decision

to treat `Dthat' in ``Logic of Demonstratives'' as an operator whose argument is a
denite description [ibid., 579, n. 29].) (2) From the point of view of semantics,
`That' is an autonomous, well-formed, lexically complete, singular term (like a
constant) but in need of a pragmatic interpretation on each of its occurrences. The
demonstration, like a stage direction, plays this pragmatic role, facilitating the
assignment and thereby completing the interpretation. I shall not pursue the issue
here, but related questions will arise for dthat-descriptions below.
26. Kaplan (1989a), 490; on Husserl and demonstratives, see Smith (1981),
(1982).
27. Although I have found some variation among speakers' intuitions, many nd
sentences like (5) and (7), interpreted as occurring in one context, to be not simply
false but ungrammatical. For discussion of this alternative applied to demonstra-
tives, see Braun (1996) and below. Note also that cross-contextual statements can
be informative regardless of whether the characters of their indexical elements are
the same or dierent. (4) is uninformative and trivial, as we originally observed,
only when the whole sentence occurs in one context (as one naturally assumes).
However, split between two contexts, especially ones that are remotely separated
over time, (4) might be as informative as (5).
28. Much more would obviously have to be said in defense of this as a general
strategy to explain the informativeness of utterances containing (exclusively)
proper indexicals; I oer one such defense in Stern (ms.).
29. See David Braun (1996), who proposes a second argument for positing rep-
resentational dierences in the characters of complete demonstratives that turns
simply on the truth-value, and not cognitive signicance, of demonstrative sen-
tences. Consider an utterance of (*) `That is larger than that', where the speaker
demonstrates in a single context one object for the rst occurrence of the demon-
strative and a dierent thing for the second occurrence. Since the two occurrences
of `that' have the same linguistic meaning, if linguistic meaning determines char-
acter, they must also have the same character. But one character in the same
context determines the same content. How, then, can (*) ever be true?
30. Note that Kaplan distinguishes the ``Fregean Theory of Demonstrations''
from the ``Fregean Theory of Demonstratives.''
31. Although the typical token, or performance, of a presentation is by a partic-
ular agent at a particular time and place, I also assume that the identities of the
agent, time, and place (or indeed that the presentation has an agent, time, and
place) do not enter into individuation of its type. Kaplan himself argues (1989a,
524.) that a presentation does not require an agent and that no particular per-
spective is essential to it, but he does think that the appearance of each presenta-
tion type must be presented from some perspective. Hence he takes the standard
form for a presentation (P), cited in the text below, to be ``The individual that has
appearance A from here now,'' adding the last two indexicals. My own view is that
it is true that every appearance is seen from some perspective (and that every pic-
ture depicts from some perspective), but the representation of that appearance
(type) need not include a representation of the perspective from which it is seen, or
336 Notes to pages 98102

of the fact that it is seen from a perspective. Hence there is no need to explicitly
mention the time and place from which the presentatum is presented, using the
indexicals `here' and `now'. Furthermore, inclusion of the indexicals `here' and
`now' in the character of a presentation-type creates a technical problem for
Kaplan's solution to Frege's puzzle for demonstratives. Specically, it can be
shown in the Logic of Demonstratives that, where (and only where) they contain
indexical or demonstrative constituents, there exist models in which two demon-
strations q and b have dierent characters while their respective complete
demonstratives have the same character. For details of this argument, see Stern
(1979), ch. 4.
32. The superscript Ps in (9) are presentation-quotes: The quotation-name names
the presentation described by the enclosed denite description.
33. To be more precise, this is true of the Fregean position as reconstructed in a
possible worlds, or points of evaluation, framework like Kaplan's Logic of
Demonstratives. The reconstruction is not, however, unproblematic. As Gilead
Bar-Elli (pers. comm.) has noted to me, the Kaplanian reformulation analyzes
sense (content) as an intension, or function from possible worlds to extensions; for
Frege, however, functions belong to the realm of referents, not senses. Thus the
reformulation obscures a fundamental Fregean distinction. For further diculties
with the analysis of content and character in terms of functions on points of
evaluation, see Braun (1995).
34. This supposition is not without problems. If, as Kaplan suggests in the quo-
tation, we think of an appearance as picture-like, pictures are both syntactically
and semantically distinct from descriptions. For discussion of these dierences,
see, e.g., Goodman (1976) and below, ch. 7, sec. IX. Kaplan alludes to the fact
that he is aware of these dierences and of the consequent impossibility of
``translating'' a picture into a description when he says that we ``try''at most try,
I would addto put the picture into words.
35. A similar strategy can be used to solve Braun's problem, raised above in n. 29,
posed by multiple occurrences of a simple demonstrative as in `that is larger than
that'.
36. This treatment is not without its problems, e.g., those mentioned in n. 31 and
n. 34. On other disanalogies between presentations and descriptions, see Stern
(ms.). Nonetheless the Fregean theory of demonstrations is on the right track in its
approach to the problem of cognitive signicance for complete demonstratives,
and it is superior to other accounts, such as Kaplan's indexical theory of demon-
strations.
37. Where F is already rigid or directly referential, this eect of `Dthat' will be
redundant.
38. See Kaplan's own statements in his (1989b), 581582 that testify to the pres-
sures pulling him in opposite directions.
39. Two other arguments oered by Kaplan for the simplicity of dthat-
descriptions bear mention. First, he argues that `` `dthat' was intended to be a
surrogate for a true demonstrative, and the description which completes it was
intended to be a surrogate for the completing demonstration,'' to which he adds
Notes to pages 103106 337

parenthetically: ``a `pointing,' being extralinguistic, could hardly be a part of syn-

tax'' (1989b, 581). But even if one is skeptical about the possibility of incorporat-
ing pointings and gestures into syntax, the relevant notion of demonstration is not
the gesture but the presentation, and, on the Fregean theory of demonstrations,
presentations can be assigned a syntax. Second, although Kaplan's main reason
for denying that `Dthat' is an operator is to avoid the conclusion that the Dthat-
description is propositionally or content-wise complex, once we distinguish charac-
ter from content, it is not obvious why syntax should mirror content rather than
character. Dierences in character, as much as in content, can make a dier-
ence in ``semantical form''so long as the latter is not identied with proposi-
tional form. Indeed, given Kaplan's own identication of characters with the
bearers of cognitive signicance and logical truthin contrast to content, which
is the bearer of metaphysical properties like necessity and contingencyone
would expect syntax to display the semantic structure of character and not that of
content.
40. Further support that the characters of dthat-descriptions are syntactically
structured is to be found by comparison with complex demonstratives of the form
`that F'. For discussion of similarities and dierences between the two classes of
expressions, see Neale (1993); Richard (1993b); Higginbotham (1988); Davies
(1982b); Taylor (1980); and Braun (1994).
41. The dilemma has its origins, perhaps, in Kaplan's original invention of
`Dthat' in his (1989b) as an attempt to capture the semantically signicant features
of Donnellan's idea of the referential use of a denite description. One way out of
the dilemma would be to treat the dthat-description as lexically complex but syn-
tactically simple. (Cf. Richard 1986.) Let the lexicon contain a rule for generating
expressionse.g., if F is a primitive expression in L, then `Dthat[F]' is alsoeach
of which would count as a primitive for purposes of syntax and semantics, i.e., as
an expression whose semantic value is not a function of the values of its parts and
their mode of construction. But there are a number of diculties with this solu-
tion. (1) It is implausible to deny that dthat-descriptions have constituent syntactic
structure. Co-referential (and co-extensive) terms can be intersubstituted within
their scope preserving the reference (extension) of the whole, and transparency is
the very mark of constituent syntactic structure. (Note that the extensionality of
dthat-descriptions follows from the fact that their direct referent is always the
individual actually denoted by the constituent description. Despite their exten-
sionality, dierent factors legislate against the possibility of quantifying into
dthat-descriptions.) (2) The possible worlds semantics framework makes it dicult
if not impossible to mark this distinction, given the provable equivalence of the
characters of `Dthat[F]' and `The x: AN[Fx]'and the undeniable fact that the
latter is syntactically complex. One might try to gloss this second problem by
conceding the limitations of the formal system. And in fact these conicting ele-
ments are the result of an attempt to construct a lexical representation to express a
use of language, thereby creating new representational structure not already pres-
ent in the expression whose use is being modeled. However, glossing the diculty
does not solve it.
42. I am indebted here to Mark McCullagh for this analogy and its moral.
338 Notes to pages 107115

Chapter 4
1. On nominative metaphors, which I argue are derived from predicative meta-
phors, see below, ch. 6, sec. VI. As with predicate demonstratives (see ch. 3, sec.
III), I also want to extend the idea of direct reference from singular terms inter-
preted metaphorically to general terms interpreted metaphorically. On the idea
that metaphors may ``directly'' refer to or express properties ( just as singular
demonstratives directly refer to individuals), see below, ch. 5, sec. VI.
2. See Black's comment that such an interpretation would ``produce an eect of
paradox and provoke a demand for justication'' (1962, 40)i.e., baement or
incomprehension. Black also notes the community-relative status of associated
commonplaces, and hence of metaphorical interpretations. However, his exam-
ple``men who take wolves to be reincarnations of dead humans will give the
statement `man is a wolf ' '' (ibid.) a dierent interpretationis not clear. Does he
mean that, with such a belief, `man is a wolf ' would metaphorically mean that
(each?) wolf is a reincarnated human?
3. Searle (1993), 105; on the role of ``stereotypes'' in meaning, see Putnam (1975).
4. See Crimmins (1992a), 9496. Crimmins refers to normal notions and ideas,
rather than features, where notions and ideas are mental particulars; my use of his
term is not meant to bear his metaphysical commitments. Indeed I am not care-
fully distinguishing features as attributes of things and as notions or ideas of those
attributes. In any case, the normal/idiosyncratic distinction is sociological, not
metaphysical: Both kinds of features are publicly shareable, accessible to more
than one individual to entertain, and mind-independent. Nor is there one reason
or explanation for the ``normalcy'' of all ideas.
5. Rubinstein (1972), 9091. The article is rich in many such examples.
6. As Black (1962) already noted; the properties, he writes, ``may, in suitable
cases, consist of deviant implications established ad hoc by the writer'' (78).
7. See White (1996), 175177, to whom I owe both the example and its explication.
8. D'Avanzo (1967), 15; cited in White (1996), 305, n. 9. D'Avanzo is comment-
ing on a letter from Keats to Reynolds of February 19, 1818.
9. For a suggestive survey of types of examples and their respective theories, see
Thompson and Thompson (1987).
10. See Stalnaker (1972); (1973), 450; (1974); (1978). Strictly speaking, Stalnaker
denes the context set as the set of possible worlds in which the presuppositions
are true, not as the presuppositions themselves. For our purposes, we need not
commit ourselves to a stand on his particular formal representation of presuppo-
sitions; nor need we commit ourselves to Stalnaker's possible worlds analysis
of propositions. Cf. also Soames (1982), Kartunnen (1974), and Gazdar (1979),
315332.
11. Strictly speaking, presupposition is a relation between an agent and a propo-
sition. Here I speak as if the contextual parameter consists of presupposed
properties rather than propositions. More precisely, what is presupposed is the
proposition that certain properties are m-associated with the literal vehicle for the
metaphor, the expression F that is interpreted metaphorically. We can, however,
Notes to pages 116121 339

dene property-presupposition in terms of the primary notion of presupposition

whose object is a proposition: If P is a property and Q a proposition, then for any
P, any speaker S, and any individual x, P is presupposed by S of x i there is some
Q containing . . . Px . . . such that Q is presupposed by S. Note that this is not
intended as a characterization specically of metaphorically relevant property-
presupposition; i.e., presuppositions involving specically properties presupposed
to be m-associated with the expression under interpretation. For discussion of
this condition, see Leezenberg (1995), 169.
12. On the semantic notion, see Strawson (1952), 175.; van Fraassen (1978). The
semantic conception of presupposition was troubled by a number of problems that
led to the rise of the pragmatic notion: e.g., the unclear relation between presup-
position and entailment, the diculty distinguishing conventionally and con-
versationally induced presuppositions, and the projection problem. Although the
pragmatic notion rejects logical entailment as the dening source of presuppo-
sitions, it retains the insight of the semantic notion that presupposition is a func-
tion of the expressions used on the occasion.
13. The pragmatic and semantic notions are not, then, incompatible. That a
presupposition is semantically required may be one reason it is pragmatically
required for the appropriateness of the utterance, although not all pragmatically
necessary presuppositions are semantically necessary.
14. See Stalnaker (1974), 201202 on barbershop small talk where the partici-
pants pretend not to know the obvious, or so represent themselves, simply in order
``to be civil, and to pass the time.''
15. Turbayne (1970), 13. See also Hunter (1973), 519: ``A metaphor pretends that
one thing is something else, thus making an implicit comparison between two
things.'' It is also possible that these views assume the awed literal deviance
thesis.
16. See Stalnaker (1978).
17. Soames (1982), 430431.
18. Stalnaker (1978), 322323. This is also an example of the process by which an
assertion can change a context, which Lewis (1979) refers to as accommodation.
19. Black (1993), 3641; cf. Goodman (1976), 7879; (1984), 68.
20. Manns (1977), 174.
21. Contrast this sense of `making us see or notice' with Davidson's causal ac-
count of metaphorical eects in WMM, as discussed earlier in ch. 2, sec. III.
22. In this case, we might say that the speaker did not use his metaphor ``teleo-
logically,'' i.e., as directed to express that particular content. At the other extreme,
one might imagine a purely ``attributive'' (or blind) use of a metaphor to express
the content, whatever it might be, that is determined in its context by the character
of the metaphor. Most cases of metaphor are, I would think, mixed. The speaker
has a content in mind that he uses the metaphor to express, but he is also willing
to allow that he will have expressed, or that he will be committed to, whatever
additional properties are presented and generated by the same metaphorical
character. I return to these issues in chs. 5 and 7.
340 Notes to pages 122134

23. Stalnaker (1973), 449.

24. Cf. Caton (1981).
25. Cf. Soames (1982) on ``taking p for granted.''
26. Stalnaker (1973), 451. Stalnaker explicates the relation between the utterance
and presupposition (and the force of ``would'') in part by arguing that presuppo-
sition is a linguistic or speech disposition. Because of independent worries about
the explanatory power of dispositional accounts of language or speech, and the
lack of an underlying mechanism for the disposition, I instead focus on the nomic
(or conventionally law-like) character of the relation. Both the dispositional and
nomic analyses support counterfactual or subjunctive conditional claims about the
relation between the utterance and its presuppositions.
27. Here is another way of making the point. The pragmatic notion of presuppo-
sition characterizes it both as a propositional attitude and as that which is required
for the appropriateness of an utterance. Prima facie we'd say, as does (P2), that p
is a presupposition of an utterance u of a sentence s in its context c because u
would be inappropriate in c unless its respective speaker S actually believed, or
represented himself as believing, that p. Instead the claim is that its respective
individual speaker S presupposes that p whenever he utters s (which requires p
to be appropriate) because p is a presupposition of s (which is the case, in turn,
because any normal speaker of s would presuppose that p). This reverses the order
of explanation.
28. See Stalnaker (1973), 451.
29. Strictly speaking (8) is a simile but, as I shall argue in ch. 6, secs. VIIVIII,
similes should be analyzed on the same model as metaphors. Hence the dierence
makes no dierence for our purposes.
30. This is one of various kinds of mixed metaphor; for discussion of others, see
ch. 5. It should be noted that not all mixed metaphors are inappropriate. A true
master of metaphor is able to exploit the prima facie inappropriateness of a mixed
metaphor to yield a powerful combination. For some examples, see White (1996),
137144.
31. In addition, it may not be clear what it means to say that the state is a car.
In terms of the notion of exemplication, which I'll develop in ch. 5, it is not
obvious that cars (unlike ships) exemplify any property attributable to states or
governments.
32. This is another example of Lewis's (1979) idea of contextual accommodation.
33. Of course, the claim is not that all metaphorical statements, without excep-
tion, have determinate truth-conditions. Like some literal statements, some meta-
phorical statements also do not have determinate contentand for the same
variety of possible reasons: vagueness, incompleteness, ambiguity, and so on. For
my view, it is strictly speaking sucient if there is at least one metaphor with a
determinate content.
34. In a Chomskian grammar (circa GB), character roughly corresponds to the
semantic level of ``logical form (LF),'' i.e., those aspects of its meaning that are
determined by sentence-grammar. Strictly speaking, LF is narrower and more
Notes to pages 135137 341

nely individuated than character since two co-intensional expressions of constant

character have the same character but possibly not the same LF. In any case, LFs,
like characters, do not by themselves determine content, truth-conditions, or
``semantic representations'' (SRs), in Chomsky's terminology. They do, however,
in combination with contextual features or the products of other ``cognitive fac-
ulties.'' See Chomsky (1977), 163.
35. On this example, see White's (1996), 4549, excellent discussion. White
describes such cases as ``ambiguity of construal.'' In a similar vein, he discusses
Davidson's example: `Tolstoy was a great moralizing infant', which he correctly
points out admits multiple interpretations. With the apparatus of my theory, we
can distinguish the dierent interpretations by the dierent scopes of the Mthat-
operator, e.g.,
(a) Tolstoy Mthat[`was a great moralizing infant']
(b) Tolstoy Mthat[`was a']i great moralizing Mthat[`infant']i .
(b) might be equivalently stated as
(b*) Tolstoy Mthat[`was an infant'] and a great moralizer
(c) Tolstoy Mthat[`was a great']i moralizing Mthat[`infant']i .
(c) might be equivalently stated as
(c*) Tolstoy Mthat[`was a great infant'] and a moralizer.
In (b) and (c) I have used subscripts to ``link'' the supercially separate con-
stituents of the metaphor whose interpretations are dependent on each other. It is
an open question whether all such possible interpretations can in fact be realized
or whether there are constraints that mark one reading and exclude others. On the
issue of individuation of metaphors, see ch. 5, sec. III.
36. On this model of metaphor recognition, see ch. 1, sec. I. The sense in which all
the alternative characters are made simultaneously available is that there is no
xed ordering among them, according to which one character in the metaphor set
(such as (14a), the so-called literal character) must be initially entertained and
then excluded prior to entertaining alternatives. So, since we can ignore all such
ordering, for all theoretical purposes we can think of the alternatives as simulta-
neously available.
37. Austin (1962), 142147. For an Austin-inspired approach to semantics, com-
patible with my argument, see Recanati (1996). I am also indebted to Dan Sperber
for raising some of these issues in conversation. However, the dierence between
the second and third stages is grounds not to assimilate metaphor simply to ``loose
talk,'' as Sperber and Wilson (1985/86) propose.
38. Note that in both the literal and metaphorical cases, it would be a mistake
to argue that simply because the sentence uttered in one context is called ``true''
and in another context ``false,'' we should assign dierent contents, or truth-
conditions, to the utterances.
39. I am not claiming, I should add, that the truth-predicate is itself relativized.
The context-dependent ``roughness'' of an absolute truth-predicate will be mir-
rored by a parallel context-dependent ``roughness'' in the right-hand clause of the
Tarskian biconditional that gives its truth-condition.
342 Notes to pages 138147

40. See Evans (1984), Maimonides (1963), Strauss (1952), and ch. 8 below.
41. For a helpful description in similar terms of this process of interpretation, see
also White (1996), 8084.
42. On exemplication, see below, ch. 5, secs. IIIII.
43. I emphasize assertion, not because it is the only or the most important use of
metaphor, but because it is presently the best understood of the many uses of
language. However, it should be emphasized that the very same principles that
apply elsewhere in explaining language use apply here; there is no need to appeal
to any rules specic to metaphor. Cf. Sperber and Wilson (1995), 237, for a similar
moral drawn from their not entirely dierent explanation of metaphor in terms of
relevance.
44. On ``Quinean,'' see Fodor (1983), 107. Of course, even with the dierence
highlighted in this paragraph, the p- and f-presuppositions are far from the kind of
input employed by an informationally encapsulated module. If nothing less than
the latter is required for semantics, it is not clear that any stage of metaphorical
interpretation will pass the test.
45. Very few predicates in the language are this straightforward, with an explicit
one- or two-criterion ``denition.'' The predicate may also have other contents or
meanings, but I assume that this is uncontroversially one of them.

Chapter 5
1. For examples of metaphor that cannot be neatly forced into the mold of simi-
larity, see Miller (1993) and, in reply, Fogelin (1988).
2. Here I take the terms of similarity to be the referent of the literal vehicle of the
metaphor, (namely, the sun) and the referent of the subject term of the sentence
(namely, Juliet). See, however, Black (1962, 35) who seems to hold (confusingly)
that the relation holds between the metaphorical expression M and its ``literal
equivalent'' L.
3. This clarication disposes of Searle's argument that, since similarity statements
entail that each of the terms of the similarity relation possess the shared property,
the comparison theory cannot account for metaphors that express properties that
are true of the subject of the statement but only believed to be true of the referent
of the vehicle of the metaphor (or, even worse, are known to be false of the refer-
ent, though they are part of the stereotype of the term). Searle charges that the
similarity theory should (contrary to fact) count all such metaphors false. For a
dierent reply, see Fogelin (1988), 4345. On my view that dierent metaphors
may indeed have dierent grounds, the ground of these ``stereotypical'' metaphors
is not similarity at all but features in the stereotypical characterization. Hence
these counterexamples do not necessarily count against the similarity thesis as a
ground for some (but not all) metaphors.
4. On the primacy of metaphorical predications, see below, ch. 6, sec. VI. Some
authors recognize the distinction between similarity as a ground of metaphor and
the logical form of metaphors but have no way of articulating it; see, e.g., Skulsky
(1986). Most psychologists of metaphor fail to attend to the distinction at all,
although, in all fairness, they are more concerned with how we understand or
Notes to pages 147149 343

comprehend metaphorsi.e., the processes we usethan with what is being pro-

cessed, the nature of the knowledge or competence, be it content or character, the
so-called task.
5. For replies to many objections to similarity theories of metaphor, see Fogelin
(1988), 3368.
6. The author of the example is R. P. Blackmur, cited by Beardsley (1962/1981),
110.
7. Fogelin (1988), 62, who calls this widely cited objection the ``reversibility
argument,'' credits it originally to Monroe Beardsley; for a recent statement of the
objection, see Moran (1989), 93, who states (erroneously, as I shall argue next):
``resemblance and similarity are both symmetrical relations: If A resembles B, then
B resembles A. Hence, if metaphor were some kind of assertion of resemblance,
we should be able to reverse any of the parts without loss or change of meaning.''
As Beardsley (1962/1981) originally formulated the objection, it was aimed
against the claim that metaphors are (elliptic) statements of similarity or compar-
ison (or elliptic similes). However, the objection applies even apart from that ad-
ditional assumption. It should also be noted that Beardsley's formulation of the
``reversal'' is not true to form: The ``reversal'' of `this man is a lion' is not `this lion
is a man' but `a lion is this man', just as the reversal of `this man is like a lion'
would be `a lion is like this man'. The whole noun phrase, not just its head, must
be reversed. The same confusion over reversibility runs through the literature; see,
e.g., Glucksberg and Keysar (1990).
8. The asymmetry of similarity statements was, interestingly, already noted by the
rabbis of the Talmud. The fth-century Babylonian sage Rava argues that a day
of rain is ``greater than the day of the Mosaic revelation'' based on the scriptural
prooftext ``My discourse [i.e., the Mosaic revelation] shall come down like the
rain'' (Deut. 32, 2). Rava explains: in a similarity statement, one term is ``depen-
dent on'' or subordinate to the other: The small one (i.e., the less important, more
peripheral case) is dependent on the great one (i.e., the prototype or central case).
Here, then, rain, which occupies the predicative position, must be more central,
important, or prototypical than the Mosaic revelation, which is in the subject
position, thus subordinate to what lls the predicative position. See BT Ta`anit 7a.
9. Similarity is a special case of a ``matching function'' that measures the degree
to which two objects, represented as sets of features, match each other depending
on measures of their common and distinctive features, weighted according to their
relative salience and according to their relative importance. Formally, the match-
ing function f for the similarity of a and b [sa; b] can be stated as
(MF) s(a; b) f (A X B, AB, BA)
or, more specically, as the linear combination, or contrast,
(CFM) s(a; b) Ff (A X B)qf (AB)-bf (BA)
where the function f is a measure of the salience of the features in A and B; F, q,
and b are parameters marking the relative importance of the shared and distinc-
tive features; A X B are the features shared by A and B; and (XY) are the fea-
tures distinctive of X.
344 Notes to pages 150156

10. See Fogelin (1988). For an instructive, and perhaps the earliest, attempt to
apply Tversky's work to metaphor, see Kittay (1982).
11. Formally: s(a; b) Ff B (A X B)qf A (AB)bf B (BA). Where the feature
has either high salience in both A and B (in which case, it will be judged that a
and b are literally similar) or low salience in both (in which case, the two will be
judged not similar), Tversky's and Ortony's similarity equations make the same
predictions.
12. A second criterion of metaphoricity for Ortony is attribute inequality, the fact
that in many metaphors or nonliteral similarity statements the A and B features
are drawn from dierent domains. It cannot, then, be (literally) the same or the
identical attribute that is common to the subject and referent; at best similar
attributes are shared or matched. Attribute inequality, Ortony suggests, either
enhances the salience imbalance or sometimes contributes to judgments of meta-
phoricity in place of increased salience of the B features (Ortony 1979, 168169).
13. In emphasizing the direct role of salience in their account, I am not disagree-
ing with G&K's own stress on the structure of metaphors as class-inclusion state-
ments. With that claim I entirely agree, although I take it to be a matter of the
logical form of metaphors rather than the nature of the groundand that is my
concern here. For the most recent version of their account, see now Glucksberg (in
press).
14. G&K (1990) do not refer, as I do, to inference to mark the dierence. They
write that the one can be ``paraphrased'' as the other or that the metaphorical
predication is ``available'' (7) for metaphorical but not literal comparisons. I
would argue that, for truth-relevant purposes, the two are equivalent and that the
function of `like' is sometimes presuppositional or aspectual. For reasons of space,
I cannot pursue this here. However the disanalogy is described, the dierence is
signicant.
15. See Peirce (1955), 102 and Goodman (1976), 52. For a not unrelated idea, see
also Henle (1958) and Alston (1964), (1980).
16. A few comments: (1) As indicated in the text, I depart from Goodman whose
nominalism forces him to take exemplication to be reference to labels rather than
properties. For discussion of diculties with his nominalism in this connection,
see Stern (1988); I assume that properties, if not unproblematic, are unavoidable.
(2) Goodman introduces exemplication to account specically (though not
exclusively) for the symbolic and referential properties of works of art; my use of
the notion is not meant to imply that all metaphors are works of art. (3) Although
Goodman discusses both exemplication and metaphor, it has been only recently
that he has attempted to put them together; see his (1984), 6165.
17. See also Goodman's refusal to supply ``general instructions how to determine
what a work exemplies'' (Goodman 1978, 172).
18. On the signicance for metaphor of the display involved in exemplication,
see below, ch. 7.
19. Note that this is not to deny an underlying resemblance relation. Insofar as
the metaphor is true and all things possess the properties they exemplify, Juliet
Notes to pages 157170 345

and the sun will resemble each other in having P. However, this similarity is a
consequence rather than a ground of the metaphorical interpretation.
20. It goes without saying that this analysis of the passage is meant merely to
illustrate how our account applies to an actual example; more sensitive and better-
informed readers are invited to oer better explications of the text.
21. Goodman (1984), 83.
22. For yet another sun-metaphor, see the example from Antony and Cleopatra,
discussed earlier in ch. 5, sec. I and in White (1996), 175177.
23. See Brooke-Rose (1958), 209.
24. Wittgenstein (1980), 15.
25. One dierence between the two is that for exemplication metaphors the
exemplied properties must be true of the exemplifying object; not so for features
in normal notions, some of which (e.g., stereotypical features) may be false of the
literal referent.
26. Goodman (1976), 7172. Although my idea of exemplication/sampling is
adopted from Goodman, he did not, at least initially, connect his idea of family-
transfer with the schema-relativity of sampling. For later thoughts associating the
two, see Goodman (1984), 6369.
27. Cf. White (1996), 8083 on this example.
28. See ch. 1, sec. III (i); ch. 4, n. 35.
29. See White (1996), 43. on this example. Note that `great' is an attributive
adjective; hence it cannot be interpreted, even metaphorically, independently of a
(explicit or implied) noun it modies.
30. On this example, see again White (1996), 58f.
31. For further discussion of this metaphor see ch. 7 below. Although it is not
obviously based on exemplication, it is an extended metaphor; for further dis-
cussion, see section IV below. The question-marks prefacing the two metaphorical
expressions in l12 mark my uncertainty whether these should be identied as
metaphors.
32. One might consider this passage a mixed metaphor but, again, with the pro-
viso that not all mixed metaphors are awkward or unacceptable. Here there is no
interpretive diculty moving from one to the other metaphor, in part because they
are interpretively independent.
33. See Grandy (1987) and Kittay and Lehrer (1981).
34. See Kittay and Lehrer (1981), 3163 (reprinted as ch. 7, 1 of Kittay 1987);
Tirrell (1989), Lako and Johnson (1980), Lako and Turner (1989), Sweetser
(1990), and White (1996), among others.
35. The other main determinant of the meaning of a word in semantic eld theory
is its set of paradigmatic relations: its relations to other words, either like or unlike
in meaning, with which it can co-occur in the same linguistic context, hence that
belong to its same grammatical category; e.g., synonyms, antonyms, hyponyms,
converses, parts and wholes, and contraries. Exemplication-schemas consisting
346 Notes to pages 170172

of contraries are one example of paradigmatically related words, although our

schemas can vary with context more than the standard cases of paradigmatically
related words. In any case, the description of paradigmatic relations, and the elds
they partition, as ``semantic'' is not straightforward. Even where the paradigmatic
relations are relatively stable across utterances, and where knowledge of the par-
titioned lexical groups is shared within the linguistic community, knowledge of
this is not specic to the language faculty, unlike syntax and semantic character.
Therefore, without drawing a sharp analytic-synthetic distinction, it is not clear
how to distinguish such knowledge from other community-wide empirical back-
ground knowledge or presuppositions. For this reason, I treat this knowledge as
contextual but utterance-invariant. Metaphorical interpretations based on utter-
ance-invariant contextual knowledge also remain relatively constant across utter-
ances; their speakers ``carry'' their knowledge of these lexical elds with them
from utterance to utterance, furnishing them with ready-made schemas to inter-
pret certain metaphors exemplicationally. On these ``routinized'' metaphors, see
below, ch. 8, sec. III.
36. In Aspects-style transformational grammar, thematic relations were expressed
by the subcategorization rules of the base; see Chomsky (1965), 90106. On the
connection between Kittay's syntagmatic relations and thematic roles, see also
Ludlow (1991).
37. There has been relatively little specic discussion of verbs interpreted meta-
phorically, but see Torreano (1997), who follows Glucksberg and Keysar's (1990)
model of metaphorical comprehension. She does not, however, discuss the role of
thematic roles in metaphorical interpretation.
38. The verb might also take an instrument, and indeed it is not impossible that in
(10) the prepositional phrase is instrumental rather than locational. This dierence
does not, however, aect my argument.
39. I owe some of these examples to Margalit and Goldblum (1995). See also
White's (1996, 9092) critical presentation of Beardsley's (1962, 295) criticisms of
John Crowe Ransom, who, following I. A. Richards's terminology, had proposed
that the underlying vehicle in Shakespeare's metaphor `Mark how the blood of
Caesar follow'd it/ As rushing out of doors . . .' (Julius Caesar III, ii, 178) is a page
rushing out of doors, shifting away from the blood, which is (in Richards's sense)
the tenor. Beardsley attacks Crowe (and Richards) for their ``idiosyncratic''
choice, or invention, of a page who is obviously not to be found in the actual lines
of the poem. White uses this discussion to motivate his own idea that metaphori-
cal interpretations are underdetermined by the meanings of their words; that is to
say, in our terminology, that the characters of the expressions interpreted meta-
phorically, or metaphorical expressions, do not, independently of their extra-
linguistic context (i.e., presuppositions) determine their content (in a context). I
would add that both Crowe and Beardsley seem to be right in part. Crowe cor-
rectly sees that the metaphor `rushing out of doors' requires an Agent as thematic
role, hence some person, such as a page. Beardsley correctly sees that this re-
quirement does not uniquely determine a single (kind of ) person (or noun phrase
referring to a person) to discharge the thematic role. To that degree, `page' as
opposed to `householder' is undetermined and, if you will, idiosyncratic. But
Notes to pages 173185 347

Beardsley fails to mention that some person (or noun phrase) is necessary. Insofar
as `page' is sucient to play the role and the role must be discharged, the choice is
not idiosyncratic.
40. Kenner (1959), 87; cited by Brooke-Rose (1958), 206. Note also Brooke-
Rose's comment that the verb metaphor ``only changes a noun implicitly. As
Professor Kenner says, `we aren't calling a ship a plough' '' (ibid., 207). On the
contrary, this is precisely how the verb metaphor determines its thematic argu-
ment. Cf. also White (1996), 272, n. 31 and Nowottny (1962), 5657 for similar
readings.
41. Cf. Jackendo (1983).
42. See Lako and Johnson (1980), Lako (1987), Lako and Turner (1989),
Turner (1987), Sweetser (1990), and Lako (1993).
43. For some brief comments, see Stern (1982) and, for a judicious critical review,
M. S. McGlone's chapter in Glucksberg (in press).
44. See Jackendo and Aaron (1991), 324.
45. I am glossing over many dierences between our accounts concerning the
semantic status of metaphorical interpretation. Rather than say that we sub-
stantively disagree, it would be more correct to say that we address dierent kinds
of phenomena involving metaphor. Terminology aside, their concern is with issues
I regard as pragmatic: issues that bear on the presuppositions that ground partic-
ular metaphors in context. Unlike Lako, I see no incompatibility between the
role of networks in the interpretation of metaphors and classical semantic theory
(within which I include my theory).
46. My own ``metaphorical expressions'' (of the form `Mthat[F]') are, of course,
terms of art.
47. G&K use ``prototype'' or ``prototypical'' where I use ``exemplar.'' In the psy-
chological literature, the former refers to feature-representations relevant to the
evaluation of typicality judgments about categories, the latter to perceived indi-
viduals that are members of categories. To avoid theoretical complications with
the former, and to emphasize the parallel to exemplication where the exemplify-
ing object is an individual, I have shifted to the latter term. Note also that when
G&K write, e.g., that ``to refer to someone as a Demjanjuk alludes to the original
war criminal, and also makes metaphorical use of his name'' (1993, 410) or that
``metaphor vehicles can have two referents simultaneously'' (ibid., 412), they are
unsuccessfully attempting, for lack of theoretical vocabulary, to distinguish the
two kinds of reference involved in exemplication. Similarly, their failure to dis-
tinguish the question of the logical form of metaphors from the question of the
role of similarity and exemplication in the generation of metaphorical properties
leaves it very unclear how they would analyze, say, verbs used metaphorically. In
order to recast them in the class-inclusion form, they would run into the same
kinds of problems that face comparison theorists. For a brief discussion, see now
Glucksberg (in press), ch. 3. However, if we distinguish logical form from grounds
like exemplication, we can claim, on the one hand, that verb-metaphors also
exemplify or sample certain actions or events (types) and, on the other, that their
(one-place) predicative logical form is an independent matter.
348 Notes to pages 186190

48. See Lako (1993), 231; cf. Lako and Turner (1989).
49. So ``exemplication'' is the answer to Lako 's question: ``How is it possible
for one kind of thing (a general situation) to be metaphorically categorized in
terms of a fundamentally spatial notion like `conning' '' (1993, 236)?
50. A similar sort of loose generalization of the term ``metaphor'' to include
all mappings is to be found in Lako 's use of the ACTIONS ARE SELF-
PROPELLED MOVEMENTS metaphor in step 3. The metaphorical force of this
mapping emerges only when one takes into account the destination (thematic)
argument of ``movements'' that metaphorically expresses the purpose of an action
(e.g., making progress toward one's end is moving forward). Once one drops this
element of the thematic relation, it is not clear that much that is metaphorical
remains in the mapping.
51. By the same token, G&K do not always suciently acknowledge the role of
systematic networks (like those described by Lako ). For example, they admit
that `has a crumbling foundation' in the sentence `the theory's foundation is
crumbling' does not itself refer to a superordinate category; it only ``implicitly
acknowledges'' that theories belong to the superordinate categories of struc-
tures (G&K 1990, 420). But without something like networks, the question arises
of how one identies relevant or appropriate but only implied superordinate
categories.
52. Lako may assume that the SPECIFIC is physical (rather than psychological)
because of his narrow conception of the nonmetaphorical or literal: namely,
``those concepts that are not comprehended via conceptual metaphor'' (1993, 205)
and instead are understood directly by way of our ``physical experiences,'' our
sensorimotor experience of space, body and orientation, social relations, and our
``culture'' (a notion Lako never explains; see Lako and Johnson 1980, 56.;
Lako 1993, 78.) Again, he writes: ``As soon as one gets away from concrete
physical experience and starts talking about abstractions or emotions, metaphori-
cal understanding is the norm,'' suggesting that the metaphorical/nonmetaphorical
distinction is to be drawn along the lines of these dierent subject-matters or
domains and, in particular, that the subject matter of the literal is the physical or
sensory. If metaphorical mappings are at bottom mappings from the non-
metaphorical, that may also be why Lako assumes that something physical (or
known by sensorimotor abilities and skills) is necessarily meant by `jail' in its
literal meaning. For reasons of space, I shall not discuss Lako 's conception of
the literal. For some discussion of its empiricist and associationist orientation, see
Jackendo (1983) and Jackendo and Aaron (1991). Indeed the deepest dierence
between Lako 's and Jackendo 's respective approaches to these issues lies in the
dierence between the kinds of cognitive mechanisms (empiricist vs. rationalist)
that they take to be employed in metaphorical interpretation.
53. Cf. also Keysar and Glucksberg (1992), 652.; Gibbs (1992b).
54. See Keysar and Glucksberg (1992), 651653; Gibbs (1994); and Glucksberg
(in press), especially ch. 5 by M. S. McGlone.
55. Here I use the word ``catachresis'' as does Black (1962), 33, n. 8, without its
negative connotations of misuse.
Notes to pages 191206 349

56. For discussion of this use of metaphor in science, see Boyd (1993).
57. Compare G&K's statement that in ordinary languages that have names for
superordinate categories, the category referred to by a metaphor, e.g., `a jail' in
`my job is a jail' ``can be described by a list of distinguishing features but it is dif-
cult to enumerate these features exhaustively'' (1990, 410). The point is not just
that it is ``dicult'' to enumerate its membership conditions, but that we lack
sucient understanding to do so.
58. Two other prima facie analogous problems of applied metaphor might be
mentioned here. One is the use of metaphors to describe the expressive properties
of works of art. It is widely acknowledged that its metaphorical character is
essential to artistic expression; Goodman indeed makes metaphorical exempli-
cation a necessary condition of expression. On the other hand, it is also assumed
that expression-claims, e.g., `the painting is sad [expresses sadness]', are truth-
valued; hence that they have propositional content, and so what they say ought to
be equally expressible in literal language. Although this problem appears similar
to the case of theological metaphors, I am not convinced they ought to be treated
the same way. Unfortunately, I do not yet have an explanation for this problem.
A second case is the use of metaphors to describe phenomenologically charac-
terized psychological states, especially in literature; on this topic, see Denham
(1998).
59. Alston (1980), 139. The argument presented here is more or less Alston's
although I have made no attempt to be faithful to all details of his exposition.
60. For a similar kind of argument, see my discussion of Max Black's cognitive
autonomy theory of metaphor in ch. 7. There I argue that it is possible to capture
the distinctive metaphorical information by character-istic information. Depend-
ing on one's theology, that may also be possible for theological metaphors.
61. Strictly speaking, no creature or created thing could exemplify these prop-
erties, since exemplication implies possession. The ``pointing'' must be less direct.
62. On the individuation of properties expressed by metaphors, see ch. 6, sec. III.
63. Pascal (1966), 112 (no. 270); cf. also 105 (no. 255), and 205 (no. 302).

Chapter 6
1. The qualication is not vacuous; for an argument that there are grammatical
constraints on metaphorical interpretation, see Stern (1983).
2. For simplicity, where F consists of the copula `is' followed by a denite de-
scription (e.g., `is the sun'), I treat it as a one-place predicate. I also assume that
names in predicative position function like predicates rather than singular terms.
Therefore, in `Quayle is no Kennedy', I take the proper name to be predicative, as
in `is a Kennedy', and to express a property as its propositional content. I shall
return below to proper names in nominative position where the story is dierent.
3. I owe the phrase to Cohen (1990). My four kinds of incompetence are not
exhaustive; Cohen assures me that there are always 613 ways to get things wrong.
4. Kaplan (1979), 402403. The rst approach is usually attributed to Montague-
Scott, the second to Frege.
350 Notes to pages 207218

5. I am indebted to Michael Byrd, many years ago, for the example, and to
Charles Parsons, also many years ago, for discussion of the subject of these para-
graphs. On metaphors and counterfactuals, see also Tormey (1983) and Margalit
and Goldblum (1995).
6. See Stalnaker (1968), (1976); and for a dierent but related account (to which I
owe the idea of contraction) Levi (1977). Nothing in my solution depends on a
particular theoretical account of counterfactuals.
7. There has been virtually no discussion of the behavior of metaphors in indirect
discourse or the attitudes. One exception is L. J. Cohen (1993) who argues that,
because metaphorical interpretation is ``preserved'' from oratio recta to oratio
obliqua, a metaphor cannot be a speech act (``metaphorizing'') and it cannot be
explained by speech act theory. He does not, however, discuss the problems dis-
cussed below that arise in moving from direct to indirect discourse. For criticism
of Cohen, see Lamarque (1982), Martinich (1984), and also White (1996), 188
189. Although I agree with White's desideratum that the content of the metaphor
in the belief- or indirect-discourse report must be the same as its content when it
occurs in its original (reported) assertive utterance, I see no reason to conclude
that all such reports are ``decidedly peculiar, or jokey'' or ``highly perverse'' (188).
8. See Kaplan (1989a), 555, n. 71 on the pseudo de re.
9. I ignore for now whether a report in other words (than those of the original
utterance), with a dierent character, but that nonetheless expresses the same
content, suces for a true report. I'll assume that the same general position,
whatever it turns out to be, that applies to the question of how far a literal report
can deviate from the reported utterance will apply to metaphor.
10. See Kaplan (1989a), 510512, on ``monsters,'' operators whose operands are
characters and shift the context to which the characters apply. Cf. also Richard
(1982) and Salmon (1986), 2744, for examples of tense operators where it does
appear that some interpretations must be relativized to contexts other than the
speaker's; and Recanati (1996, 1997) for extended discussion of the problem of
context-shifting. Despite Recanati's qualications of Kaplan's theory, none of his
counterexamples violates the general principle because of which Kaplan bans
monsters.
11. Where the speaker already has similar enough, if not the same, presup-
positions as the subject of his report, their interpretations of the metaphor will, of
course, ipso facto be the same or close enough. This is not infrequently the case,
and not only with more or less dead metaphors. See my discussion in ch. 8 of
``routinized metaphors'' that have live, productive characters that apply to their
contexts to yield their respective contents, but that ``carry'' their contexts along
with them from utterance to utterance.
12. The idea of insulated presuppositions is analogous to Barwise and Cooper's
(1981) notion of nonpersistence. A nonpersistent statement is a statement that
holds in a situation s but possibly not in a larger situation s 0 extending s. A set of
presuppositions cs is insulated i there is some assertion s that is interpretable in cs
but not in a larger set of presuppositions cs 0 that contains cs as a proper part.
Notes to pages 219231 351

13. Ibid. Here Fogelin assumes that metaphors are gurative comparisons or
likenesses; hence his reference to likeness claims.
14. On the semantics of `good' and similar adjectives, with special attention to
their context-dependence, see Kamp (1975) and Bartsch (1987).
15. See Fogelin (1988), 91, n. 9. Fogelin ignores the fact that where the truth-
values of utterances of the same sentence dier, so must their truth-conditions
or propositional contents. He (correctly) wishes to sever shift of reference from
shift of meaning, but does not allow for the possibility of shift of propositional
content.
16. The literal and metaphorical contents of the expression (under its respective
interpretations) are dierent properties, but they are not dierent kinds of content.
Nor are the two kinds of truthmetaphorical truth (the truth of an utterance
under its metaphorical interpretation) and literal truth (the truth of an utterance
under its literal interpretation)dierent kinds of truth (whatever that would
mean).
17. See Washington (1992) and Carpintero (1994). If the value is not identical to
the displayed token, it may be the type of that token.
18. Jakobson (1960).
19. Searle (1993); cf. Fogelin (1988), 38.
20. Throughout this section I am indebted to conversation with Patti Nogales,
who rst pointed out to me Fogelin's problematic passage. See her discussion of
the argument in Nogales (1999), 183185.
21. On metaphorical reference, see Berg (ms.).
22. See Kaplan (1973) on blind and directed pointings.
23. For a dierent view of the relation, see White (1996), 234245.
24. Fogelin (1988), 28, proposes that ``Aristotle seems to be using the term `meta-
phor' in a broad generic sense as a way of indicating that similes are also gures
of speech,'' i.e., that ``metaphor'' here refers to the whole class of gures. I know
of no other passage in Aristotle where he unequivocally uses ``metaphor'' in this
broad sense, and, other than the fact that this reading supports Fogelin's own
view, it is also far from the best interpretation of the passage in question. There-
fore, I'll stick with the narrow meaning of ``metaphor'' and what Aristotle seems
to be literally, and insightfully, arguing.
25. For classic expressions of this view, see Cicero (1942) 3.38.15639.157
and Quintilian (1922) Bk. VIII, vi, 89, who explicitly say that metaphors are
``shorter'' similes. Among recent authors, the best defense of the (sophisticated)
position is Fogelin (1988).
26. For reasons that will immediately become clear, this claim is not identical
to the stated claim in ch. 5, although it captures the correct idea behind the
formulation.
27. Like any assertion, that of a metaphor ``makes'' presuppositions, among
them, those necessary for its interpretability in its context including the
352 Notes to pages 232241

comparison-judgment through which the property that is metaphorically ex-

pressed is determined. In this sense, we might say that metaphorical utterances
`make' comparisons, not that those comparisons are either a causal eect of the
utterances or their contents, what they assert.
28. On Fogelin's use of ``elliptic,'' see also his explanation (1988), 45; clearly it is
not syntactic or linguistic ellipsis. On ``like'' as a hedge, see Lako (1973).
29. See Goodman (1976) and Fogelin (1988), who both assume that there is one
class of gures. Note that the gures or tropes do not exhaust the forms of non-
literal interpretation, such as ellipsis, e.g., Dylan Thomas's `a grief ago' that
should probably be interpreted as elliptical for `a period of grief ago', or `the sun is
very hot today' (pers. comm., James McCawley) that should be interpreted as `the
air heated by the sun is hot today'.
30. Cohen (1975), 670.
31. Grice (1975), 71. My italics.
32. Searle (1993), 108109. My italics.
33. Fogelin (1988), 910.
34. On complex gurative statements, see Berg (1988), who does not develop the
following argument but to whose imaginative examples I am indebted for sug-
gesting the argument to me.
35. Further evidence that irony is postsemantic is the strong eect of intonation
and tone on irony. Irony is also often taken to be an expression of feeling, attitude,
or evaluationunlike metaphor that is not associated with a specic feeling or
attitude. On the dierence between irony and metaphor, see also Winner (1988),
3234; Winner and Gardner (1993); Gibbs (1993); and Sperber and Wilson (1981).
36. Following my practice throughout this book, I do not include under the
heading ``semantic'' the pragmatic, use, or speech act theories of Searle, Fogelin,
Grice, Cooper, Ted Cohen, and others. Because of our very dierent conceptions
of semantics, I also don't attempt to do justice to the older generation of theories
like those of Black, Beardsley, Henle, or Richards. No one, of course, has a
monopoly on the term ``semantics'' but the issue is the character of the explana-
tory theory, and I assume that this dierence will be acknowledged even by those
who insist that the aforementioned theories ought to be called ``semantic.''
37. Matthews (1971); Levin (1977); and to some extent Beardsley's later versions
of his theory in (1976), (1978).
38. See Beardsley (1978); also Katz (1977), 1322; Levin (1977); Mack (1975);
Matthews (1971).
39. Cohen and Margalit (1972), 735; see also Cohen (1993) for a later version of
the same idea.
40. On the virtues of ``the method of cancellation'' as opposed to ``the method of
multiplication'' in semantic theory more generally, see Cohen (1993), 6164.
41. Cohen (1993) claims that ``lexical entries for a natural language can draw no
clear distinction between features that are supposed to be `purely linguistic' and
features that are supposed to represent common knowledge or commonly
Notes to pages 241247 353

accepted beliefs'' (75); as evidence, he cites the lexicographical practices of the

Concise Oxford Dictionary. Although I am sympathetic to worries about a sharp
analytic-synthetic distinction, it is not obvious that it parallels the dictionary-
encyclopedia distinction or that dictionary entries should be taken as authoritative
evidence of the mental lexicon insofar as it is part of linguistic competence.
42. I am indebted here to Judy Feldmann for raising this issue.
43. Kittay (1987); all in-text parenthetic page references for the remainder of this
section are to this work. For an introduction to semantic eld theory, see Lehrer
(1974) and Grandy (1987).
44. See Kittay's repeated statements that her semantic features include empirical
and pragmatic considerations, e.g., 50, 5657, 8182, 91, 140. On the semantic
status of semantic eld theory in general, see Grandy (1987).
45. At the end of her account, Kittay herself seems to acknowledge that not all
metaphors in fact do involve some kind of incongruity, even pragmatic incongru-
ity. To account for the non-incongruous examples, she proposes another principle
(or principles, the ``general'' and ``specic independence of applicability condi-
tions'' principles) according to which an utterance is ambiguous, or polysemous,
with its literal and metaphorical interpretations. It remains unclear to me why,
having made this explicit concession, Kittay continues to assert that incongruity is
a necessary condition for being a metaphor.
46. The example originates in Reddy (1969); see also Beardsley (1978) and, for
criticisms of Beardsley similar to those I level against Kittay, Stern (1983).
47. It should be noted that linguistic, or semantic, anaphora is entirely intra-
sentential or sentence-internal. Although there is a rich literature on the linguistic
constraints on anaphoric relations, especially on disjoint anaphoric reference, we
have relatively little theoretical understanding of how cross-sentential anaphora
works. The two should not be assimilated to one model.
48. Two further problems with Kittay's analysis should be mentioned. (1) Even
though she claims that literal incongruity is a necessary condition for recognition
and interpretation of a metaphor, Kittay acknowledges that it is not sucient.
What, then, would be sucient (given incongruity) on her account? Kittay pro-
poses the following ``alternative conditions'': ``if the utterance was intended to be
understood metaphorically or if it is possible to attribute a metaphorical interpre-
tationappropriate to the contextto the utterance'' (148). It is dicult to see
how these criteria do not beg the question. (3) In cases of truly novel metaphor,
Kittay supposes that the topic domain may be unarticulated, at least conceptually;
instead of conceptual units, the terms for the constituents in the topic domain
must represent ``relatively discrete experiential phenomena''(170). Exactly how the
Expressibility Principle can make these nonconceptual phenomena verbally ex-
plicit requires considerably more explanation that Kittay oers. Apart from the
question of how one can map one semantic eld onto a second if it is non-
conceptualized, it is also hard to see how there can be incongruity or contradiction
between two elds, one of which is unarticulated and nonconceptual.
49. See Grandy (1987) and Kittay (1987), 230257.
354 Notes to pages 248260

50. Cf. Kittay (1987), 171, 175176, 233.

51. In some passages Kittay seems to suggest that the point of her project is not,
as I have been assuming, to show how it is possible to supplement our received
conception of semantics with semantic eld theory in order to construct a theory
of metaphor, but to jettison the received conception. In reply to Davidson, she
counters: ``Second order discourse [which includes metaphor] also has a deep
structure . . . which will not submit to the logical form of rst order discourse,''
and ``truth-theoretic semantics . . . will give the logical form or deep structure only
of sentences interpreted in their rst-order meaning'' (141142). These remarks
suggest that a theory of second-order meaning like metaphor must reject, or
replace, truth-theoretic semantics (at least for that purpose).
52. Of course, the fact that Kittay can formalize the function (see 144145, 172
175) does not itself show that the function is semantic, nor does the formalization
per se explain it.
53. I am indebted here to Jesse Prinz for discussion.
54. On this, see below, ch. 7, sec. III.
55. Cf. Bergmann (1982).
56. On this recurring theme in theoretical linguistics, see Chomsky (1965), (1980),
and Higginbotham (1982).
57. Of course, there is the obvious dierence that knowledge of the contextual
parameters for the indexicals (the time, place, speaker, etc.) is generally shared
and accessible to all ordinary speakers, whereas knowledge of the presuppositions
necessary for metaphorical interpretation may, as Aristotle noted, sometimes
require ``genius.'' Likewise, there are obvious dierences between individuals'
capacities to produce metaphors and their capacities to comprehend them. But all
this is a matter of knowledge of context rather than knowledge of metaphorical
character.
58. I am indebted here, and throughout this reply, to extremely helpful comments
by Roger White.

Chapter 7
1. The example originates with Merleau-Ponty, but I learned of it from Danto
(1978). Alternative accounts that do not take `swallow' to be metaphorical might
exploit the causal powers of the sound or phonetic properties of the word, or take
it as a pun or clang-word in Freud's terminology. In stipulating that (1) is a meta-
phor, I do not wish to dismiss but merely to bracket the identication question.
2. A similar example, reported in Rubenstein (1972), 92, concerns a writer ``who
gave ample evidence of the presence of intense unconscious castration anxiety.''
After submitting an article to a prestigious magazine, he was asked by the editor
to cut his article, a request to which the writer reacted with inordinate rage.
Rubinstein hypothesizes that the writer ``read the letter [of the editor] as request-
ing him not to cut (i.e., shorten) the article but to cut (i.e., to cut o ) the
ARTICLE (i.e., his penis and/or testicles).'' Thus his violent reaction was in per-
fect accord with his fear.
Notes to pages 260270 355

3. For a dierent but not incompatible interpretation of this passage, see T.

Cohen (1997).
4. Cf. also Wayne Booth's ``catchsh'' example in Booth (1978).
5. Ibn Ezra (1975); cited in Septimus (ms.). The English translation is Septimus'.
On Ibn Ezra's theory of metaphor, see now Fenton (1997).
6. Houseman (1933), quoted and criticized in Brooks (1965b).
7. Vickers (1988), 743.
8. See Hobbes (n.d.) I, 5; Locke (1975), III, X, 34.
9. For a similar argument, see Davidson (1984), 261.
10. The phrase ``cognitive autonomy'' was coined, I believe, by Margalit (1970).
Proponents include Richards (1936), Brooks (1965b), and Black (1962).
11. Another argument against literal paraphrasability as a requirement on all
metaphors would focus on the de re metaphors we discussed in ch. 5, sec. VI.
12. Cf. Bergmann (1982).
13. On this motif, see Burge (1979b), (1989).
14. Cavell (1967), 7980, where he also quotes Empson.
15. I owe the phrase to Margalit and Goldblum (1994).
16. I do not mean to suggest that there do not remain serious problems as to how
to individuate contexts and contents. If we identify the context of a metaphor with
a set of presuppositions, strictly speaking two contexts are the same if and only if
they contain the very same presuppositions; yet one context can be an extension of
another and, in practice, certain dierences are frequently ignored. Similar prob-
lems infect the idea of ``one interpretation.'' Thus when literary theorists cite
examples of poetic metaphors that have admitted, or borne, repeatedly new
``interpretations'' over time as evidence for the claim that a (good?) metaphor is
inexhaustiblethat each time we read it, we nd something new in itit is not
absolutely clear that these cases should be described as ones in which the meta-
phor has one, always extendible, or endless, interpretation in one context or dif-
ferent, though related, interpretations, each of which is nite, relative to dierent
contexts of interpretation. For a further obscurity, see my later discussion in sec-
tion V of the problems in distinguishing the presemantic presuppositions that
determine the schema to which a metaphor belongs from the semantically relevant
presuppositions about the features sorted by the schema.
17. I borrow the term ``directions'' from Strawson (1950); cf. White (1996), 40
41.
18. Cf. Boyd (1993) and White (1996), 180, 307308 (to whom I also owe the
term ``embryonic'' meaning).
19. As Jim Higginbotham once observed to me (pers. corr.), if it were true that,
for each paraphrase of a metaphor, we could always explicitly identify yet another
feature not part of the paraphrase, then why not just tack that feature onto the
paraphrase? And if we can't make it explicit, why think that the paraphrase as it
stands is not complete?
356 Notes to pages 271286

20. For an example of how the Gricean conversational maxims that govern
assertion accomplish this, see above, ch. 4, sec. IV.
21. Cf. White (1996), 4041.
22. Or so it was reported in Time Magazine, February 18, 1985.
23. For this example, I am indebted to Ted Cohen who assures me that he only
mentions but never uses it. Both the Nigerian `bedbug' and American (or W. C.
Fieldsian) English `chickadee' are, I assume, partly dead metaphors although I
also assume that they are both still recognized as metaphors, i.e., their inter-
pretations are context-dependent and ``computed'' from metaphorically relevant,
culturally xed presuppositions. It is also worth observing, along the lines of our
remarks in ch. 1, sec. III (iv), that, for our present purposes, namely, to illustrate a
xed, fairly determinate content, such partly dead metaphors are preferable to
completely ``live'' ones whose interpretation would require a great deal of con-
textual stage-setting in order to x their content determinately.
24. Chomsky (1959).
25. See Scheer (1991); the terminology is his.
26. See, however, Scheer (1991), 12, n. 23. Note also that, if one insists on
conict, the prior beliefs or expectations (e.g., the belief that [literally] Juliet is not
the sun) will also generally be tacit.
27. What of so-called twice-true metaphors that involve no anomaly? For an
example like Ted Cohen's `no man is an island', one might argue that its inter-
pretation proceeds by negating what we would understand by the metaphorical
interpretation of an utterance of `a man is an island'. The surprise would then be
linked to the simple atomic statement rather than its negation, along the lines
suggested in the text. For utterances like the tautologous `men are men', possibly
it is also the antecedent unlikelihood that we could ever say something true and
informative about the content of the one expression using the other that makes
the metaphorical interpretation surprisingif it is. In either case, whatever the
exact explanation, it seems clear that it must revolve around the character of the
metaphor.
28. It is not certain that this network is based on presuppositions involving a
symbolic relation like exemplication; it may be historically linked to the not-
much-earlier invention of printing and the growing power and elitism based on
literacy.
29. On this passage, see also Tirrell (1989) and Thompson and Thompson (1987),
169f. On the book-metaphor elsewhere in Shakespeare, see also White (1996),
65. and above, ch. 5, sec. III.
30. Cf. Tirrell (1989) on the ``expressive commitment'' of a metaphor: a ``com-
mitment to the viability and value of a particular way of talking about something''
(22), and Moran (1989) on ``framing.''
31. Cf. also Ricoeur (1978), 144.
32. Likewise, Davidson (1986) denies that there is an answer to the question
``How many facts or propositions are conveyed by [say] a photograph?'' because
this is a ``bad question. A picture is not worth a thousand words, or any other
Notes to pages 286292 357

number. Words are the wrong currency to exchange for a picture''or, for the
same reason, for a metaphor (263).
33. Moran (1989), 94101; it is not clear whether Moran himself endorses the
conclusion of the argument or simply formulates it to motivate Davidson's move.
Davidson hints at the argument when he denies, for example, that the aspects and
resemblances that a metaphor makes us see are ``denite cognitive content that its
author wishes to convey and that the interpreter must grasp if he is to get the
message'' (1986, 262). The term ``irresistible'' is borrowed from Wayne Booth
(1978), 54.
34. Moran's target is more specic than my presentation indicates. He takes the
irresistibility of metaphor to be a problem specically for those who both think of
metaphor paradigmatically as successful metaphor, where the success involves
``framing'' together the terms of the metaphor, and also hold that metaphors make
assertions.
35. The fact that metaphors and pictures ``bring'' an indenite, even potentially
innite, number of things ``to our attention'' should count no more against their
possession of content or meaning than the fact that the most literal of utterances
can likewise bring such an unbounded number of things to our attention counts
against their having a meaning or content. Nor is it clear why the fact that even
the simplest of pictures typically has multiple truth-conditionsi.e., the fact that
its ``truth-conditions'' do not correspond one-to-one to the truth-condition of a
single atomic sentenceshould count against its having truth-conditions or prop-
ositional content, as Davidson seems to suggest. For a similar, and similarly
obscure, argument in a dierent context, see Fodor (1975), 180.
36. See Thompson and Thompson (1987), 163165.
37. Davidson (1986), 253, my italics. Cf. also the critical remarks on this passage
by Thompson and Thompson (1987), 175176.
38. Indeed there is some evidence that metaphors that play exclusively on purely
visual appearances, e.g., `the moon is a sickle', are those that are least connected
to a family. So, to the extent to which a metaphor is based just on a visual prop-
erty, the less pictorial it is in our sense, where the pictorialness of the metaphor is
due to its place in a network.
39. Goodman (1976), 130232; Haugeland (1981); see also Peacocke (1986),
(1989) and, for a thorough overview, Rollins (1989).
40. For further discussion, see Stern (1997). Throughout the last four paragraphs,
I am indebted to discussion and correspondence with David Hills. As I noted in
ch. 3, n. 34, this dierence between pictures and discursive symbols or descriptions
has signicant implications for Kaplan's use of descriptions as a model for non-
linguistic presentations (which he thinks of as something like pictures), hence for
his attempt to use dthat-descriptions as a model for complete demonstratives. For
further discussion, see Stern (forthcoming). It is this dierence that possibly lies
behind Kaplan's (1989a) comment that we can only ``try to put the appearance
into words'' (526). The same dierence, I should note, legislates against certain
analyses of pictorial metaphors, such as Goodman (1976), that require replication
for transfer; on this, see below, ch. 8, sec. I and Stern (1997).
358 Notes to pages 292296

41. One should not conclude that pictures are necessarily more ``nely'' individ-
uated than descriptions. ``Fineness'' may vary along dierent dimensions. For an
argument that the characters of two co-denoting descriptions (in a digital schema)
might dier while visual presentations (that would belong to an analog/pictorial
schema) of the co-referent would not, see Peacocke (1989), 306312.
42. I am indebted for this bit of scholarship to W. R. Johnson and Christopher
Bobonich. Another, more contemporary, example would employ `Superman' and
`Clark Kent' as metaphors (ignoring problems of reference of ctional names).
43. These limitations of the notion of character clearly reect the same general
limitations of the notion of character for the semantics of proper names; see
above, ch. 3. Furthermore, to the extent that intuitive, philosophical dierences
like this are not expressed by our formal apparatus, this is a sample of the limi-
tations of formalization in semantics; on this theme, see Kaplan (1989a,b).
44. See also our third criterion for the individuation of metaphors in ch. 5, sec.
III, according to which a metaphor is individuated by its character, content, and
context (including network). Although the linguistic type of the metaphor is pre-
sumably individuated by its character, to capture its full cognitive signicance a
ner criterion like the third is necessary.
45. Cavell (1967), 79. For a similar suggestion, see also Moran (1989), 112. I
should add that, as with complete demonstratives that can have richer and
poorer presentational contents, there is a continuum of dierences of degree
to which metaphors are sensitive to their respective schemas and, correspondingly,
a continuum of dierences in their respective character-dependent cognitive
signicances.
46. This is the qualifying complication to which I alluded earlier in section IX.
Furthermore, in Stern (forthcoming), I argue that a similar situation obtains for
the complete demonstrative whose character is a function of the character of
its presentation component, which is also pictorial, i.e., dense, analog, or non-
discursive. Hence our knowledge of neither the character of a complete demon-
strative nor that of a metaphorical expression is knowledge only of descriptional,
propositional language. But if we are willing to grant complete demonstratives a
kind of knowledge or cognitive value associated with their character, metaphors
deserve equal treatment.
47. Note that, despite his ocial hard line, Davidson (1978) himself admits in his
unguarded moments that ``of course [what we notice or see by a metaphor] may be
[propositional] and, when it is, it usually may be stated in fairly plain words'' (263,
his italics). See also White's (1996) discussion of Davidson's claim (194203). I
agree with White that the propositional content of the metaphor in its context
roughly, what he suggests would be captured by the ``School Comprehension
Test''does not ``capture the creative signicance of the use of metaphor'' (201),
but I would argue that we can begin to capture it by taking into account the per-
spective carried by its metaphorical character (relative to its networks) and that
such additional signicance in no way excludes the propositionality of metaphor-
ical content. I am indebted here, too, to correspondence with White.
Notes to pages 297304 359

48. Cf. Crimmins (1992a), 44., and Richard (1990). Note that Crimmins's
counterexamples are of dierent ``ways of thinking'' associated with proper
names. Since the characters and contents of co-directly referring proper names
(and other simple eternal expressions) are equivalent, it is not surprising that there
is no systematic individuation of ways of thinking by their characters or linguistic
types. Indeed Crimmins's counterexamples are really no more than a good illus-
tration of the inadequacy of Kaplan's character/content-based semantic theory to
account for proper names or other simple eternal expressions.
49. Although it is not possible to individuate arbitrary ways of thinking by char-
acters or linguistic types, it may be possible to do so for specic kinds. For ex-
ample, it may be possible to appeal to Crimmins's (1992a) idea of normal notions
associated with, or individuated by, the characters of some expressions, relative
to linguistic communities, to account for the special role of the indexicals in
explaining human actions. Associated with the rst-person indexical `I' would be a
normal notion of an agent that consists (among other things) of a set of dis-
positions to act in certain ways in certain circumstances; associated with `now'
would be a normal notion that consists of a set of dispositions to perform actions
that their agents believe should be performed at certain times at those same times
(or at times believed to be the same as those times). Note that if it is a normal
notion that is associated with, and individuated by, the character of the indexical
that does the explanatory work, the indexical will be just as essential as it would
be if it were its character proper doing the explanation. Of course, individual
speakers may associate their own idiosyncratic notions with individual tokens or
occurrences of the indexicals in addition to the normal notion.
50. Searle (1993), 102.

Chapter 8
1. Lako (1993), 241.
2. See especially Goodman (1984), 5570.
3. See Gibbs (1984), (1989); Dascal (1987), (1989); Recanati (1995).
4. See Johnson (1981), 171177.
5. By the ``merely possible'' I mean the nonactual possible since, on most theories
of possibility, the actual (and necessary) is also possible. Depending on one's
theory, the ctional may belong here as well. I take ``actual'' and ``real'' to be
synonymous (as well as ``actually,'' ``really,'' and ``in fact'') and the present to be
the temporal analogue of the actual.
6. Here I treat ``actual'' as an indexical; see Lewis (1978); and Kaplan (1989a). On
the actual truth of metaphors, see also Goodman (1976), 6880.
7. This argument, and understanding of the literal, also rests on another confu-
sion that assimilates literal language to the language of empirical descriptions,
histories, and scientic theories, and metaphors to the language of ctions, myths,
and other imaginative texts. See, e.g., Cassirer (1946), 8399. For a more sophis-
ticated account of metaphor as a kind of ``ctional language,'' see Walton (1993).
360 Notes to pages 305310

Works of the former kind are in turn assumed to be, in some clearly recognized
but hard-to-make-precise sense, ``about'' reality, the real world, actuality, or what
is true in fact. Fictions, myths, and other products of the imagination are not
``about'' reality or actuality in the same way, and those who mistakenly interpret
them as if they were descriptions of the actual or real world should, in turn, be
criticized for taking them literally or ``at face value'' (Margalit and Halbertal
1992, 84). This division is confused. Metaphors are not themselves works like
myths, ctions, or imaginative texts, although those works may employ meta-
phors. But they may also (and obviously do) employ literal language, and they are
no less mythic or ctional or imaginative if they are written in literal language
from start to nish. On the other hand, scientic descriptions and explanations of
the actual world, as well as histories, also make frequent use of metaphors, and
they are no less truth-valued for their metaphorical mode of expression. In sum,
the literal-metaphorical distinction cannot be collapsed onto distinctions between
the scientic and the mythic/ctional or between the actual and imaginative.
8. We also nd recognition of the literal meaning of a text among the Greeks, as
well as dierences between levels of meaning or interpretation, say, in Philo and
among the Neo-Platonist exegetes. See Lamberton (1986) for references. Among
Jewish exegetes, there is arguably no notion specically of the literal, although the
idea of peshat is sometimes presented as if it were literal meaning. However, a
more accurate explication of the term would be ``contextual meaning''; see Kamin
(1986) and Halivni (1991) and references therein. Throughout this discussion I
ignore many dierences of detail among medieval authors. For more detailed
discussion, see Evans (1984).
9. Evans (1984), 107, 110.
10. Evans (1984), 107.
11. See Margalit (1978); Partee (1984).
12. The qualier ``actual'' is meant to rule out cases, like some examples Searle
cites, in which the physical world is imagined to be dierent than it actually is.
Since I do not believe that any of us have clear intuitions about what we would
say or mean under those circumstances, I think it makes good methodological
sense to limit ourselves to the actual.
13. Similar remarks apply to the various deconstructionist philosophers and lit-
erary theorists who deny the literal-nonliteral distinction on the grounds that all
meaning is relative to a culture or various kinds of beliefs and presuppositions.
True as this observation surely is, all it shows is that all meaning is presemanti-
cally context-dependent; the literal-nonliteral distinction, which hinges on seman-
tic and postsemantic context-independence, is an entirely separate matter.
14. The examples are Henle's (1958), 186.
15. See Asch and Nerlove (1960); Winner and Gardner (1978).
16. Contrast the account of children's metaphors in Cohen and Margalit (1972),
723. Although I cannot discuss children's metaphors at length here, suce it to
say that the empirical facts are not unambiguous. One must distinguish both mis-
taken overgeneralizations from metaphorical uses and adult observers' projections
Notes to pages 311319 361

of their own metaphorical interpretations onto the children's uses from the chil-
dren's uses themselves. On projections, see below.
17. See Alston (1964), 99.
18. Compare our earlier distinction in ch. 3, sec. VII between the de facto context-
sensitivity of dthat-descriptions and the de jure context-sensitivity of indexicals.
19. I owe this phrase to Charles Parsons.
20. Alston (1964), 99103; on the continuum picture, see also Lycan (1984), 200.
21. Compare Kaplan's (1989a), 560562, idea of dubbing a new word in the
language: We announce: ``Let That[D] be A'' where there is also a change of
character.
22. This would be more in line with our conception of meaning as constraint
i.e., as character. However, the traditional notion of meaning did not distinguish
character and content, and once we do distinguish them, it not evident that it must
be character rather than content that is literal. Clearly ordinary language is no
guide here.
23. See Cassirer (1946), 8399 who argues that there is a distinct kind of meta-
phorical as opposed to literal perception. On this view, the metaphorical-literal
distinction, like the a prioria posteriori distinction, would mark a dierence, say,
in our modes of justication; for an apparently similar idea, see Black (1993).
24. Rousseau (1966), 1213. Rousseau's own explanation of his claim rests on the
assumption that the function of language is primarily to express ``passions,'' and
that ideas, in addition to words, are what are transferred.
25. Ibid., 12.
References

Almog, J. (1984), ``Semantical Anthropology,'' in P. French, T. Uehling, and

H. Wettstein, eds., Midwest Studies in Philosophy IX (Minneapolis, University of
Minnesota Press): 479489.
Almog, J. (1986), ``Naming without Necessity,'' Journal of Philosophy 83: 210
242.
Alston, W. P. (1964), Philosophy of Language (Englewood Clis, N.J., Prentice
Hall).
Alston, W. P. (1980), ``Irreducible Metaphors in Theology,'' in E. T. Long, ed.,
Experience, Reason, and God (Washington D.C., The Catholic University of
America Press): 129148.
Aristotle (1984), Poetics and Rhetoric, in J. Barnes, ed., The Complete Works of
Aristotle, 2 vols. (Princeton, N.J., Princeton University Press).
Asch, S. (1961), ``The Metaphor: A Psychological Inquiry,'' in M. Henle, ed.,
Documents of Gestalt Psychology (Berkeley, University of California Press): 86
94.
Asch, S. and Nerlove, H. (1960), ``The Development of Double Function Terms
in Children: An Exploratory Study,'' in B. Kaplan and S. Wapner, eds., Per-
spectives in Psychological Theory (New York, International Universities Press):
4760.
Augustine (1958), On Christian Doctrine, trans. D. Robertson (Indianapolis,
Bobbs-Merrill Publishers).
Austin, J. L. (1962), How to Do Things with Words (Oxford, Clarendon Press).
Barsalou, L. W. (1983), ``Ad Hoc Categories,'' Memory and Cognition 11: 211
227.
Bartsch, R. (1987), ``Context-dependent Interpretations of Lexical Items,'' in J.
Groenendijk, D. De Jongh, and M. Stokhof, eds., Foundations of Pragmatics and
Lexical Semantics (Dordrecht, Foris Publications).
Barwise, J. and Cooper, R. (1981), ``Generalized Quantiers and Natural Lan-
guage,'' Linguistics and Philosophy 4: 159219.
Beardsley, M. (1958), Aesthetics: Problems in the Philosophy of Criticism (New
York, Hartcourt, Brace).
364 References

Beardsley, M. (1962), ``Metaphorical Twist,'' Philosophy and Phenomenological

Research 22: 293307; reprinted in Johnson (1981): 105122.
Beardsley, M. (1976), ``Metaphor and Falsity,'' The Journal of Aesthetics and Art
Criticism 35: 218222.
Beardsley, M. (1978), ``Metaphorical Senses,'' Nous 12: 316.
Berg, J. (1988), ``Metaphor, Meaning, and Interpretation,'' Journal of Pragmatics
12: 695709.
Berg, J. (ms.), ``Metaphorical Reference.''
Bergmann, M. (1979), ``Metaphor and Formal Semantics,'' Poetics 8: 213230.
Bergmann, M. (1982), ``Metaphorical Assertions,'' Philosophical Review 91: 229
242.
Binkley, T. (1976), ``On the Truth and Probity of Metaphor,'' Journal of Aes-
thetics and Art Criticism 33: 171180.
Black, M. (1962), ``Metaphor,'' in Models and Metaphors (Ithaca, N.Y., Cornell
University Press): 2547; reprinted in Johnson (1981): 6382.
Black, M. (1993), ``More about Metaphor,'' in Ortony (1993a): 1941.
Blackburn, S. (1984), Spreading the Word: Groundings in the Philosophy of Lan-
guage (Oxford, Oxford University Press).
Booth, W. (1978), ``Metaphor as Rhetoric: The Problem of Evaluation,'' Critical
Inquiry 5: 4972.
Booth, W. (1988), The Company We Keep (Chicago, University of Chicago Press).
Boyd, R. (1993), ``Metaphor and Theory Change: What is `Metaphor' a Meta-
phor For?'', in Ortony (1993a): 481532.
Braun, D. (1994), ``Structured Characters and Complex Demonstratives,'' Philo-
sophical Studies 74: 193219.
Braun, D. (1995), ``What Is Character?'' Journal of Philosophical Logic 24: 227
240.
Braun, D. (1996), ``Demonstratives and Their Linguistic Meanings,'' Nous 30:
145173.
Breen, D. (1993), Rigidity, Essence, and Identity (Ph.D. dissertation, University of
Chicago).
Brooke-Rose, C. (1958), A Grammar of Metaphor (London, Secker and War-
burg).
Brooks, C. (1947), The Well-Wrought Urn (New York, Harcourt Brace, and
World).
Brooks, C. (1965a), ``Metaphor, Paradox, and Stereotype,'' British Journal of
Aesthetics 5: 315328.
Brooks, C. (1965b), Modern Poetry and the Tradition (New York, Oxford Uni-
versity Press).
Burge, T. (1974), ``Demonstrative Constructions, Reference, and Truth,'' Journal
of Philosophy 71: 205223.
Burge, T. (1977), ``Belief De Re,'' Journal of Philosophy 74: 338362.
References 365

Burge, T. (1979a), ``Sinning against Frege,'' Philosophical Review 88: 398432.

Burge, T. (1979b), ``Individualism and the Mental,'' in P. French, T. Uehling, and
H. Wettstein, eds., Midwest Studies in Philosophy IV: Studies in Metaphysics
(Minneapolis, University of Minnesota Press): 73121.
Burge, T. (1986), ``Intellectual Norms and the Foundations of Mind,'' Journal of
Philosophy 83: 697720.
Burge, T. (1989), ``Wherein Is Language Social?'' in A. George, ed., Reections on
Chomsky (Oxford, Basil Blackwell Ltd.): 175191.
Carey, J. (1981), Donne: Life, Mind, and Art (London and Boston, Faber and
Faber).
Carnap, R. (1963), ``Autobiography,'' in P. A. Schilpp, ed., The Philosophy of
Rudolph Carnap (La Salle, Ill., Open Court): 384.
Carpintero, M. G. (1994), ``Ostensive Signs: Against the Identity Theory of Quo-
tation,'' Journal of Philosophy 91: 264275.
Cassirer, E. (1946), Language and Myth, trans. S. K. Langer (New York, Dover).
Caton, C. E. (1981), ``Stalnaker on Pragmatic Presuppositions,'' in P. Cole, ed.,
Radical Pragmatics (New York, Academic Press): 83100.
Cavell, S. (1967), ``Aesthetic Problems of Modern Philosophy,'' in M. Black, ed.,
Philosophy in America (Ithaca, Cornell University Press): 7497.
Chomsky, N. (1959), Review of Verbal Behavior, by B. F. Skinner, Language 35:
2658.
Chomsky, N. (1965), Aspects of the Theory of Syntax (Cambridge, Mass., MIT
Press).
Chomsky, N. (1968), Language and Mind, Enlarged Edition (New York, Har-
court Brace Jovanovitch).
Chomsky, N. (1977), Essays in Form and Interpretation (New York, North-
Holland).
Chomsky, N. (1980), Rules and Representations (New York, Columbia University
Press).
Chomsky, N. (1988), Language and Problems of Knowledge: the Managua Lec-
tures (Cambridge, Mass., MIT Press).
Cicero (1942), De Oratore, vol. 2, trans. H. Racham (Cambridge, Harvard Uni-
versity Press).
Cohen, G. A. (1995), Self-Ownership, Freedom, and Equality (Cambridge, Cam-
bridge University Press).
Cohen, L. J. (1985), ``A Problem about Ambiguity in Truth-Theoretic Seman-
tics,'' Analysis 45: 129135.
Cohen, L. J. (1993), ``The Semantics of Metaphor,'' in Ortony (1993a): 5870.
Cohen, L. J. and Margalit, A. (1972), ``The Role of Inductive Reasoning in the
Interpretation of Metaphor,'' in D. Davidson and G. Harman, eds., Semantics of
Natural Language (Dordrecht, Reidel): 721762.
366 References

Cohen, T. (1975), ``Figurative Language and Figurative Acts,'' Journal of Philos-

ophy 72: 669690; reprinted in Johnson (1981): 182199.
Cohen, T. (1976), ``Notes on Metaphor,'' Journal of Aesthetics and Art Criticism
34: 249259.
Cohen, T. (1978), ``Metaphor and the Cultivation of Intimacy,'' Critical Inquiry 5,
1: 312.
Cohen, T. (1983), ``Jokes,'' in E. Schaper, ed., Pleasure, Preference and Value
(Cambridge, Cambridge University Press).
Cohen, T. (1990), ``Figurative Incompetence,'' Raritan 10, 2: 3044.
Cohen, T. (1997), ``Metaphor, Feeling, and Narrative,'' Philosophy and Literature
21: 223244.
Cooper, D. (1986), Metaphor (Oxford, Oxford University Press).
Crimmins, M. (1992a), Talk about Belief (Cambridge, Mass., MIT Press).
Crimmins, M. (1992b), ``Context in the Attitudes,'' Linguistics and Philosophy 15:
185198.
Culler, J. (1981), ``The Problem of Metaphor,'' in T. E. Hope et al., eds., Lan-
guage, Meaning, and Style: Essays in Memory of Stephen Ullmann (Leeds, Leeds
University Press): 520.
Danto, A. (1978), ``Freudian Explanations and the Language of the Uncon-
scious,'' in Joseph H. Smith, ed., Psychoanalysis and Language (New Haven, Yale
University Press): 325353.
Danto, A. (1981), The Transguration of the Commonplace (Cambridge, Mass.,
Harvard University Press).
Dascal, M. (1987), ``Defending Literal Meaning,'' Cognitive Science 11: 259281.
Dascal, M. (1989), ``On the Roles of Context and Literal Meaning in Under-
standing,'' Cognitive Science 13: 253257.
D'Avanzo, M. L. (1967), Keats' Metaphors for the Poetic Imagination (Durham,
N.C., Duke University Press).
Davidson, D. (1967), ``Causal Relations,'' Journal of Philosophy 64: 691704;
reprinted in D. Davidson (1980), Essays on Actions and Events (Oxford, Oxford
University Press): 149162.
Davidson, D. (1975), ``Thought and Talk,'' in S. Guttenplan, ed., Mind and Lan-
guage (Oxford, Oxford University Press); reprinted in Davidson (1984): 155
170.
Davidson, D. (1978), ``What Metaphors Mean,'' Critical Inquiry 5: 3147;
reprinted (in revised form) in Davidson (1984): 245264.
Davidson, D. (1984), Inquiries into Truth and Interpretation (Oxford, Oxford
University Press).
Davidson, D. (1986), ``A Nice Derangement of Epitaphs,'' in LePore (1986): 433
446.
Davidson, D. (1995), ``Laws and cause,'' Dialectica 49: 263279.
References 367

Davies, M. (1982a), ``Idiom and Metaphor,'' Proceedings of the Aristotelian

Society 83: 6785.
Davies, M. (1982b), ``Individuation and the Semantics of Demonstratives,'' Jour-
nal of Philosophical Logic 11: 287310.
Davis, S. (1991), ed., Pragmatics: A Reader (Oxford, Oxford University Press).
Denham, A. (1998), ``Metaphor and Judgments of Experience,'' in European
Review of Philosophy, vol. 3, eds., R. Casati and C. Tappolet (Stanford, CSLI
Publications): 225253.
Donnellan, K. (1966), ``Reference and Denite Descriptions,'' Philosophical Re-
view 75: 281304.
Elgin, C. Z. (1983), From Reference to Reference (Indianapolis, Hackett).
Elgin, C. Z. and Scheer, I. (1987), ``Mainsprings of Metaphor,'' Journal of Phi-
losophy 84: 331335.
Evans, G. (1981), ``Understanding Demonstratives,'' in H. Parret and J. Bouve-
resse, eds., Meaning and Understanding (Berlin/New York, De Gruyter): 280
303.
Evans, G. (1982), The Varieties of Reference (Oxford/New York, Clarendon
Press).
Evans, G. R. (1984), The Language and Logic of the Bible (Cambridge, Cam-
bridge University Press).
Fay, D. and Cutler, A. (1977), ``Malapropisms and the Structure of the Mental
Lexicon,'' Linguistic Inquiry 8: 505520.
Fenton, P. B. (1997), Philosophie et Exegese dans Le Jardin De La Metaphore de
Mose Ibn `Ezra, Philosophe et Poete Andalou Du XII Siecle (Leiden, Brill).
Fernandez, J. W. (1986), Persuasions and Performances: The Play of Tropes in
Culture (Bloomington, Indiana, Indiana University Press).
Fernandez, J. W. (1991), ed., Beyond Metaphor: The Theory of Tropes in Anthro-
pology (Stanford, Cal., Stanford University Press).
Fodor, J. (1975), The Language of Thought (New York, Thomas Y. Crowell Co.).
Fodor, J. (1983), The Modularity of Mind (Cambridge, Mass., MIT Press).
Fogelin, R. (1988), Figuratively Speaking (New Haven, Yale University Press).
Fogelin, R. (1994), ``Metaphors, Similes, and Similarity,'' in Hintikka (1994): 23
39.
Frege, G. (1966), ``On Sense and Meaning,'' in P. T. Geach and M. Black, trans.
and eds., Translations from the Philosophical Writings of Gottlob Frege (Oxford,
Basil Blackwell Ltd.).
Frege, G. (1984), ``Thoughts,'' in B. McGuiness (1984), ed., G. Frege, Collected
Papers on Mathematics, Logic, and Philosophy (Oxford, Basil Blackwell Ltd.);
reprinted in Salmon and Soames (1988): 3355.
Gareld, J. (1987), ed., Modularity in Knowledge Representation and Natural
Language Understanding (Cambridge, Mass., MIT Press).
368 References

Gazdar, G. (1979), Pragmatics: Implicature, Presupposition, and Logical Form

(New York, Academic Press).
Gibbs, R. W. (1984), ``Literal Meaning and Psychological Theory,'' Cognitive
Science 8: 274304.
Gibbs, R. W. (1989), ``Understanding and Literal Meaning,'' Cognitive Science
13: 243251.
Gibbs, R. W. (1990), ``Comprehending Figurative Referential Descriptions,''
Journal of Experimental Psychology: Learning, Memory, and Cognition 16: 5666.
Gibbs, R. W. (1992a), ``When is Metaphor? The Idea of Understanding in
Theories of Metaphor,'' Poetics Today 13: 575606.
Gibbs, R. W. (1992b), ``Categorization and Metaphor Comprehension,'' Psycho-
logical Review 99: 572577.
Gibbs, R. W. (1993), ``Process and products in making sense of tropes,'' in Ortony
(1993a): 252276.
Gibbs, R. W. (1994), The Poetics of Mind (New York, Cambridge University
Press).
Glucksberg, S. (in press), Understanding Figurative Language: From Metaphors to
Idioms (Oxford, Oxford University Press).
Glucksberg, S. and Keysar, B. (1990), ``Understanding Metaphorical Compari-
sons: Beyond Similarity,'' Psychological Review 97: 318.
Glucksberg, S. and Keysar, B. (1993), ``How Metaphors Work,'' in Ortony
(1993a): 401424.
Glucksberg, S., Keysar, B., and McGlone, M. S. (1992), ``Metaphorical Under-
standing and Accessing Conceptual Schema: Reply to Gibbs (1992),'' Psychologi-
cal Review 99: 578581.
Glucksberg, S. and Manfredi, D. (forthcoming), ``Metaphoric Comparisons,'' in
C. Cacciari, ed., Similarity: Proceedings of the Workshops of the International
Center for Cognitive and Semiotic Studies (Milano, Italy, Bompiani).
Goodman, N. (1972a), ``Sense and Certainty,'' in Problems and Projects (India-
napolis, Hackett): 6068.
Goodman, N. (1972b), ``On Likeness of Meaning,'' in Problems and Projects
(Indianapolis, Hackett): 221230.
Goodman, N. (1976), Languages of Art, Second Edition (Indianapolis, Hackett).
Goodman, N. (1978), ``Reply to Beardsley,'' Erkenntnis 12: 95118; reprinted in
Goodman (1984): 8085.
Goodman, N. (1979), ``Metaphor as Moonlighting,'' Critical Inquiry 6: 125130;
reprinted in Goodman (1984): 7177.
Goodman, N. (1984), Of Mind and Other Matters (Cambridge, Mass., Harvard
University Press).
Grandy, R. (1987), ``In Defense of Semantic Fields,'' in E. LePore, ed., New
Directions in Semantics (London, Academic Press): 259280.
References 369

Grandy, R. (1990), ``Understanding and the Principle of Compositionality,''

Philosophical Perspectives 4: 557571.
Grice, H. P. (1975), ``Logic and Conversation,'' in D. Davidson and G. Harman,
eds., The Logic of Grammar (Encino, Cal., Dickenson): 6475.
Gruber, J. S. (1965/1976), Studies in Lexical Relations (Ph.D. dissertation, Mas-
sachusetts Institute of Technology; distributed by Indiana University Linguistics
Club, Bloomington, Indiana; reprinted as part of Lexical Structures in Syntax and
Semantics (Amsterdam, North-Holland Publishers).
Hacking, I. (1986), ``The Parody of Conversation,'' in LePore (1986): 447458.
Halivni, D. W. (1991), Peshat and Derash (New York, Oxford University Press).
Hankamer, J. and Sag, I. (1976), ``Deep and Surface Anaphora,'' Linguistic In-
quiry 7: 391428.
Haugeland, J. (1981), ``Analog and Analog,'' Philosophical Topics 12: 213225.
Henle, P. (1958), ``Metaphor,'' in P. Henle, ed., Language, Thought, and Culture
(Ann Arbor, Michigan, University of Michigan Press): 173195; reprinted in
Johnson (1981): 83104.
Hesse, M. (1987), ``Unfamiliar Noises II: Tropical Talk: The Myth of the Literal,''
Proceedings of the Aristotelian Society, supp. volume 61: 297311.
Higginbotham, J. (1982), ``Noam Chomsky's Linguistic Theory,'' Social Research
49, 1: 143157.
Higginbotham, J. (1988), ``Contexts, Models, and Meanings: A Note on the
Data of Semantics,'' in R. M. Kempson, ed., Mental Representations: The Inter-
face between Language and Reality (Cambridge, Cambridge University Press): 29
48.
Higginbotham, J. (1989), ``Knowledge of Reference,'' in A. George, ed., Reec-
tions on Chomsky (London, Basil Blackwell Ltd.): 159165.
Higginbotham, J. (forthcoming), ``Perspectives on Demonstratives.''
Hills, D. (1997), ``Aptness and Truth in Verbal Metaphor,'' Philosophical Topics
25, 1: 117154.
Hintikka, J. (1994), ed., Aspects of Metaphor (Dordrecht, Kluwer).
Hintikka, J. and Sandu, G. (1994), ``Metaphor and Other Kinds of Nonliteral
Meaning,'' in Hintikka (1994): 151188.
Hobbes, J. (n.d.) Leviathan, ed., M. Oakeshott (Oxford, Basil Blackwell Ltd.).
Hornstein, N. (1984), Logic as Grammar (Cambridge, Mass., MIT Press).
Hornstein, N. (1986), ``Pragmatics and Grammatical Theory,'' Proceedings of the
Annual Meeting of the Chicago Linguistic Society 22: 234247.
Housman, A. E. (1933), The Name and Nature of Poetry (Cambridge, The Uni-
versity Press).
Hrushovski, B. (1984), ``Poetic Metaphor and Frames of References, with exam-
ples from Eliot, Rilke, Mayakovsky, Mandelshtam, Pound, Creeley, Amichai, and
the New Tork Times,'' Poetics Today 5: 543.
370 References

Hunter, J. P. (1973), ed., The Norton Introduction to Literature: Poetry (New

York, Norton).
Ibn Ezra, M. (1975), Kitab al-muhadara wa'l-mudhakara (Heb: Sefer ha-`Iyyunim
veha-Diyyumim), trans. and ed., A. Halkin (Jerusalem, n.p.).
Isenberg, A. (1963), ``On Dening Metaphor,'' Journal of Philosophy 60: 609622.
Jackendo, R. (1983), Semantics and Cognition (Cambridge, Mass., MIT Press).
Jackendo, R. (1990), Semantic Structures (Cambridge, Mass., MIT Press).
Jackendo, R. and Aaron, D. (1991), Review of More Than Cool Reason: A Field
Guide to Poetic Metaphor, by G. Lako and M. Taylor, Language 67: 320338.
Jakobson, R. (1960), ``Linguistics and Poetics,'' in T. Sebeok, ed., Style in Lan-
guage (Cambridge, Mass., MIT Press): 350377.
Johnson, M. (1981), ed., Philosophical Perspectives on Metaphor (Minneapolis,
University of Minnesota Press).
Kamin, S. (1986), Rashi's Exegetical Categorization in respect to the Distinction
between Peshat and Derash (in Heb.) (Jerusalem, Magnes Press).
Kamp, J. A. W. (1975), ``Two Theories about Adjectives,'' in E. Keenan, ed.,
Formal Semantics of Natural Language (Cambridge, Cambridge University Press):
123155.
Kaplan, D. (1973), ``Bob and Carol and Ted and Alice,'' in J. Hintikka, J.
Moravcsik, and P. Suppes, eds., Approaches to Natural Language (Dordrecht,
Reidel): 490518.
Kaplan, D. (1979), ``Dthat,'' in P. French, T. Uehling, and H. Wettstein, eds.,
Contemporary Perspectives in the Philosophy of Language (Minneapolis, Univer-
sity of Minnesota Press): 383400.
Kaplan, D. (1989a), ``Demonstratives,'' in J. Almog, J. Perry, and H. Wettstein,
eds., Themes from Kaplan (Oxford, Oxford University Press): 481563.
Kaplan, D. (1989b), ``Afterthoughts,'' in J. Almog, J. Perry, and H. Wettstein,
eds., Themes from Kaplan (Oxford, Oxford University Press): 565614.
Kaplan, D. (1990), ``Words,'' Proceedings of the Aristotelian Society, suppl. vol.
64: 93119.
Kartunnen, L. (1974), ``Presupposition and Linguistic Context,'' Theoretical Lin-
guistics 1: 181194.
Katz, J. J. (1977), Propositional Structure and Illocutionary Force: A Study of the
Contribution of Sentence-Meanings to Speech-Acts (New York, Thomas Y. Cro-
well Comp.).
Kempson, R. (1977), Semantic Theory (London, Cambridge University Press).
Kenner, H. (1959), The Art of Poetry (New York, Rinehart).
Keysar, B. and Glucksberg, S. (1992), ``Metaphor and Communication,'' Poetics
Today 13: 633658.
Khatchadurian, H. (1968), ``Metaphor,'' British Journal of Aesthetics 8: 227
243.
References 371

Kittay, E. F. (1982), ``The Creation of Similarity: A Discussion of Metaphor in

Light of Tversky's Theory of Similarity,'' PSA 1: 394405.
Kittay, E. F. (1987), Metaphor: Its Cognitive Force and Linguistic Structure (Ox-
ford, Oxford University Press).
Kittay, E. F. and Lehrer, A. (1981), ``Semantic Fields and the Structure of Meta-
phor,'' Studies in Language 5: 3163.
Kripke, S. (1980), Naming and Necessity (Cambridge, Mass., Harvard University
Press).
Kronfeld, C. (1980/81), ``Novel and Conventional Metaphors: A Matter of
Methodology,'' Poetics Today 2: 1324.
Lako, G. (1973), ``Hedges: A Study of Meaning Criteria and the Logic of Fuzzy
Concepts,'' Journal of Philosophical Logic 2: 458508.
Lako, G. (1987), Women, Fire, and Other Dangerous Things (Chicago, Univer-
sity of Chicago Press).
Lako, G. (1993), ``The Contemporary Theory of Metaphor,'' in Ortony (1993a):
202251.
Lako, G. and Johnson, M. (1980), Metaphors We Live By (Chicago, University
of Chicago Press).
Lako, G. and Turner, M. (1989), More than Cool Reason: A Field Guide to
Poetic Metaphor (Chicago, University of Chicago Press).
Lamarque, P. (1982), ``Metaphor and Reported Speech: In Defence of a Prag-
matic Theory,'' Journal of Literary Semantics 11: 1418.
Lamberton, R. (1986), Homer the Theologian (Berkeley, University of California
Press).
Langer, S. K. (1942), Philosophy in a New Key (Cambridge, Mass., Harvard
University Press).
Larson, R. and Segal, G. (1995), Knowledge of Meaning: An Introduction to
Semantic Theory (Cambridge, Mass., MIT Press).
Leezenberg, M. (1995), Contexts of Metaphor (Ph.D. dissertation, Institute for
Logic, Language, and Computation, University of Amsterdam, ILLC dissertation
series).
Lehrer, A. (1974), Semantic Fields and Lexical Structure (Amsterdam, North-
Holland).
LePore, E. (1986), ed., Truth and Interpretation: Perspectives on the Philosophy of
Donald Davidson (Oxford, Basil Blackwell Ltd.).
Levi, I. (1977), ``Subjunctives, Dispositions, and Chances,'' Synthese 34: 423
455.
Levin, S. (1977), The Semantics of Metaphor (Baltimore, The Johns Hopkins
University Press).
Levin, S. (1993), ``Language, Concepts, and Worlds: Three Domains of Meta-
phor,'' in Ortony (1993a): 124135.
372 References

Levinas, E. (1990), Dicult Freedom (Baltimore, Maryland, Johns Hopkins

Press).
Lewis, D. (1970), ``General Semantics,'' Synthese 22: 1867; reprinted in Lewis
(1983), Philosophical Papers, vol. 1 (Oxford, Oxford University Press): 189229.
Lewis, D. (1978), ``Anselm and Actuality,'' Nous 4: 175188; reprinted in Lewis
(1983): 1021.
Lewis, D. (1979), ``Scorekeeping in a Language Game,'' Journal of Philosophical
Logic 8: 339359; reprinted in Lewis (1983): 233249.
Locke, J. (1975), An Essay Concerning Human Understanding, ed. P. H. Nidditch
(Oxford, Oxford University Press).
Loewenberg, I. (1975), ``Identifying Metaphors,'' Foundations of Language 12:
315338; reprinted in Johnson (1981): 154181.
Ludlow, P. (1991), Review of Metaphor: Its Cognitive Force and Linguistic Struc-
ture, by E. F. Kittay, Journal of Philosophy 88: 324330.
Ludlow, P. (1997), ed., Readings in the Philosophy of Language (Cambridge,
Mass., MIT Press).
Lycan, W. (1984), Logical Form in Natural Language (Cambridge, Mass., MIT
Press).
Lycan, W. (1987), Review of Portraying Analogy, by J. F. Ross, Linguistics and
Philosophy 10: 118.
Mac Cormac, E. R. (1985), A Cognitive Theory of Metaphor (Cambridge, Mass.,
MIT Press).
Mack, D. (1975), ``Metaphoring as Speech Act: Some Happiness Conditions for
Implicit Similes and Simple Metaphors,'' Poetics 4: 221256.
Maimonides, M. (1963), The Guide of the Perplexed, trans. S. Pines (Chicago,
University of Chicago Press).
Manns, J. (1977), ``Goodman on Metaphor,'' The Personalist 58: 173178.
Margalit, A. (1970), ``The Cognitive Status of Metaphors'' (in Heb.), (Ph.D. dis-
sertation, Hebrew University of Jerusalem).
Margalit, A. (1978), ``The Platitude Principle in Semantics,'' Erkenntnis 13: 377
395.
Margalit A. and Goldblum, N. (1994), ``Metaphors in an Open-Class Test,'' in
Hintikka (1994): 219241.
Margalit, A. and Goldblum, N. (1995), ``A Metaphor Game,'' Synthese 104: 299
323.
Margalit, A. and Halbertal, M. (1992), Idolatry (Cambridge, Mass., Harvard
University Press).
Martinich, A. P. (1984), ``A Theory for Metaphor,'' Journal of Literary Semantics
13: 3556.
Matthews, R. (1971), ``Concerning a `Linguistic Theory' of Metaphor,'' Founda-
tions of Language 7: 413425.
References 373

Miller, G. (1993), ``Images and models, similes and metaphors,'' in Ortony

(1993a): 357400.
Moran, R. (1989), ``Seeing and Believing: Metaphor, Image, and Force,'' Critical
Inquiry 16, 1: 87112.
Neale, S. (1993), ``Term Limits,'' Philosophical Perspectives 7: 89123.
Newport, E. L. and Bellugi, U. (1978), ``Linguistic Expressions of Category
Levels in a Visual-Gesture Language: A Flower is a Flower is a Flower,'' in
E. Rosch and B. B. Lloyd, eds., Cognition and Categorization (Hillsdale, N.J.,
Erlbaum).
Nogales, P. (1999), Metaphorically Speaking (Stanford, Cal., CSLI/Cambridge
University Press).
Nowottny, W. (1962), The Language Poets Use (London, Athlone Press).
Ortony, A. (1979), ``Beyond Literal Similarity,'' Psychological Review 86: 161
180.
Ortony, A. (1993a), ed., Metaphor and Thought, second edition (Cambridge,
Cambridge University Press).
Ortony, A. (1993b), ``The role of similarity in similes and metaphors,'' in Ortony
(1993a): 342356.
Partee, B. (1984), ``Compositionality,'' in F. Landman and F. Veltman, eds.,
Varieties of Formal Semantics (Dordrecht, Foris Publications): 281311.
Pascal, B. (1966), Pensees, trans. A. J. Krailsheimer (New York, Penguin Books).
Peacocke, C. (1986), ``Analogue Content,'' Proceedings of the Aristotelian Society,
suppl. vol. 60: 117.
Peacocke, C. (1987), ``Depiction,'' Philosophical Review 96: 383410.
Peacocke, C. (1989), ``Perceptual Content,'' in J. Almog, J. Perry, and H. Wett-
stein, eds., Themes from Kaplan (Oxford, Oxford University Press): 297329.
Peirce, C. S. (1955), ``Logic as Semiotic: The Theory of Signs,'' reprinted in Phil-
osophical Writings of Peirce, ed., Justus Buchler (New York, Dover).
Perry, J. (1977), ``Frege on Demonstratives,'' Philosophical Review 86: 474497.
Perry, J. (1979), ``The Problem of the Essential Indexical,'' Nous 13: 321;
reprinted in N. Salmon and S. Soames (1988), eds., Propositions and Attitudes
(Oxford, Oxford University Press): 83101.
Plimpton, G. (1976), ed., Writers at Work: The Paris Review Translations, Fourth
Series (New York, Viking).
Preminger, A. (1965), ed., The Princeton Handbook of Poetic Terms (Princeton,
N.J., Princeton University Press).
Putnam, H. (1975), ``The Meaning of `Meaning','' in H. Putnam (1975), Mind,
Language, and Reality, Philosophical Papers, vol. 2 (Cambridge, Cambridge Uni-
versity Press).
Pylyshyn, Z. W. (1993), ``Metaphorical Imprecision and the ``top-down'' research
strategy,'' in Ortony (1993): 543559.
374 References

Quine, W. v. O. (1969), ``Ontological Relativity,'' in Ontological Relativity and

Other Essays (New York, Columbia University Press).
Quine, W. v. O. (1978), ``A Postscript On Metaphor,'' Critical Inquiry 5: 161
162.
Quintilian (1922), The Institutio Oratoria of Quintilian III, trans. H. E. Butler
(London, William Heinemann).
Recanati, F. (1995), ``The Alleged Priority of Literal Interpretation,'' Cognitive
Science 19: 207232.
Recanati, F. (1996), ``Domains of Discourse,'' Linguistics and Philosophy 19: 445
475.
Recanati, F. (1997), ``Context-shifting in Metarepresentations'' (Paris, C.R.E.A.
Rapports et Documents, Rapport No. 9720, C.N.R.S.).
Reddy, M. (1969), ``A Semantic Approach to Metaphor,'' Papers from the Fifth
Regional Meeting of the Chicago Linguistics Society (Chicago, University of
Chicago-Linguistics Dept.): 240251.
Reddy, M. (1993), ``The Conduit Metaphor: A Case of Frame Conict in Our
Language About Language,'' in Ortony (1993): 137163.
Reinhart, T. (1970), ``On Understanding Poetic Metaphor,'' Poetics 5: 383402.
Reinhart, T. (1976), ``Patterns, Intuitions, and the Sense of Nonsense,'' PTL: A
Journal for Descriptive Poetics and Theory of Literature 1: 85103.
Richard, M. (1982), ``Tense, Propositions, and Meanings,'' Philosophical Studies
41: 337351.
Richard, M. (1986), ``Quotation, Grammar, and Opacity,'' Linguistics and Phi-
losophy 9: 383403.
Richard, M. (1988), ``Direct Reference and Ascriptions of Belief,'' in Salmon and
Soames (1988): 169196.
Richard, M. (1990), Propositional Attitudes: An Essay on Thoughts and How We
Ascribe Them (Cambridge, Cambridge University Press).
Richard, M. (1993a), ``Attitudes in Context,'' Linguistics and Philosophy 16: 123
148.
Richard, M. (1993b), ``Articulated Terms,'' Philosophical Perspectives 7: 207230.
Richards, I. A. (1936), The Philosophy of Rhetoric (London, Oxford University
Press).
Ricoeur, P. (1962), The Rule of Metaphor: Multidisciplinary Studies in the Cre-
ation of Meaning in Language, trans. R. Czerny (Toronto, University of Toronto
Press).
Ricoeur, P. (1978), ``The Metaphorical Process as Cognition, Imagination, and
Feeling,'' Critical Inquiry 5: 143159; reprinted in Johnson (1981): 228247.
Rollins, M. (1989), Mental Imagery: On the Limits of Cognitive Science (New
Haven, Yale University Press).
Rorty, R. (1987), ``Unfamiliar Noises I: Hesse and Davidson on Metaphor,''
Proceedings of the Aristotelian Society, supp. volume 61: 283296.
References 375

Rosch, E. (1973), ``On the internal structure of perceptual and semantic cate-
gories,'' in T. E. Moore, ed., Cognitive Development and the Acquisition of Lan-
guage (New York, Academic Press).
Ross, J. F. (1981), Portraying Analogy (Cambridge, Cambridge University Press).
Rousseau, J-J. (1966), On the Origin of Language, trans. J. H. Moran and A.
Gode (New York, Frederick Unger).
Rubinstein, B. (1972), ``On Metaphor and Related Phenomena,'' in R. Holt and
E. Peterfreund, eds., Psychoanalysis and Contemporary Science, vol. 1 (New York,
Macmillan): 70108.
Rumelhart, D. E. (1993), ``Some Problems with the Notion of Literal Meanings,''
in Ortony (1993): 7182.
Russell, B. (1956), ``Philosophy of Logical Atomism,'' in R. C. Marsh, ed., Logic
and Knowledge (London, George Allen and Unwin Ltd.): 175282.
Sadock, J. M. (1993), ``Figurative speech and linguistics,'' in Ortony (1993): 42
57.
Salmon, N. (1986), Frege's Puzzle (Cambridge, Mass., MIT Press).
Salmon, N. (1988), ``Reexivity,'' in Salmon and Soames (1988): 240274.
Salmon, N. (1990), ``A Millian Heir Rejects the Wages of Sinn,'' in C. A. Ander-
son and J. Owens, eds., Propositional Attitudes (Stanford, Cal., C.S.L.I. Lecture
Notes).
Salmon, N. and Soames, S. (1988), eds., Propositions and Attitudes (Oxford, Ox-
ford University Press).
Scheer, I. (1954), ``An Inscriptional Approach to Indirect Quotation,'' Analysis
14: 8390.
Scheer, I. (1979), Beyond the Letter (London, Routledge and Kegan Paul).
Scheer, I. (1991), ``In Praise of the Cognitive Emotions,'' in In Praise of the
Cognitive Emotions (New York, Routledge): 317.
Searle, J. (1978), ``Literal Meaning,'' Erkenntnis 13: 207224.
Searle, J. (1983), Intentionality (Cambridge, Cambridge University Press).
Searle, J. (1993), ``Metaphor,'' in Ortony (1993): 83111.
Septimus, B. (n.d.), Maimonides on Language.
Skulsky, H. (1986), ``Metaphorese,'' Nous 20: 351369; reprinted in J. Gareld
and M. Kiteley (1991), eds., Meaning and Truth (New York, Paragon House):
582598.
Smith, D. W. (1981), ``Indexical Sense and Reference,'' Synthese 49: 101127.
Smith, D. W. (1982), ``What is the Meaning of `This'?,'' Nous 16: 181208.
Snyder, J. (1980), ``Picturing Vision,'' Critical Inquiry 7, 4: 499526.
Soames, S. (1982), ``How Presuppositions Are Inherited: A Solution to the Pro-
jection Problem,'' Linguistic Inquiry 13: 483545; reprinted in Davis (1991): 428
470.
Soames, S. (1986), ``Incomplete Denite Descriptions,'' Notre Dame Journal of
Formal Logic 27, 3: 349375.
376 References

Soames, S. (1988), ``Direct Reference, Propositional Attitudes, and Semantic

Content,'' in Salmon and Soames (1988): 197239.
Sperber, D. and Wilson, D. (1981), ``Irony and the Use-Mention Distinction,'' in
P. Cole, ed., Radical Pragmatics (New York, Academic Press).
Sperber, D. and Wilson, D. (1985/86), ``Loose Talk,'' Proceedings of the Aristo-
telian Society 86: 153171; reprinted in Davis (1991): 540549.
Sperber, D. and Wilson, D. (1995), Relevance: Communication and Cognition,
Second Edition (Oxford, Basil Blackwell).
Spolsky, E. (1987/88), ``The Limits of Literal Meaning,'' New Literary History 19:
419440.
Spurgeon, C. (1952), Shakespeare's Imagery and What It Tells Us (Cambridge,
Cambridge University Press).
Stalnaker, R. (1968), ``A Theory of Conditionals,'' in N. Rescher, ed., Studies in
Logical Theory (Oxford, Blackwell Publishers): 98112.
Stalnaker, R. (1972), ``Pragmatics,'' in D. Davidson and G. Harman, eds.,
Semantics of Natural Language (Dordrecht, Reidel): 380397.
Stalnaker, R. (1973), ``Presuppositions,'' Journal of Philosophical Logic 2: 447
457.
Stalnaker, R. (1974), ``Pragmatic Presuppositions,'' in M. Munitz and P. Unger,
eds., Semantics and Philosophy (New York, New York University Press): 197
213.
Stalnaker, R. (1976), ``Indicative Conditionals,'' in A. Kasher, ed., Language in
Focus (Dordrecht, Reidel Publishing Company): 179196.
Stalnaker, R. (1978), ``Assertion,'' in P. Cole, ed., Syntax and Semantics 9: Prag-
matics (New York, Academic Press): 315332; reprinted in Davis (1991): 278
289.
Stern, J. (1979), Metaphor as Demonstrative: A Formal Semantics for Demonstra-
tives and Metaphors (Ph.D. dissertation, Columbia University).
Stern, J. (1982), Review of Contemporary Perspectives on Metaphor, ed. by M.
Johnson, Journal of Aesthetics and Art Criticism 40: 231234.
Stern, J. (1983), ``Metaphor and Grammatical Deviance,'' Nous 17: 577599.
Stern, J. (1985), ``Metaphor as Demonstrative,'' Journal of Philosophy 82: 677
710.
Stern, J. (1988), ``Metaphor without Mainsprings: A Rejoinder to Elgin and
Scheer,'' Journal of Philosophy 85: 427438.
Stern, J. (1991), ``What Metaphors Do Not Mean,'' in P. French, T. Uehling, and
H. Wettstein, eds., Midwest Studies in Philosophy, vol. XVI: Philosophy and the
Arts (South Bend, Indiana, Notre Dame University Press): 1352.
Stern, J. (1997), ``Metaphors in Pictures,'' Philosophical Topics 25, 1: 255294.
Stern, J. (1998), ``Metaphor and Philosophy of Language,'' in M. Kelly, ed.,
Oxford Encyclopedia of Aesthetics, 4 vols. (Oxford, Oxford University Press),
vol. 3: 212215.
References 377

Stern, J. (2000), ``Maimonides on Language and the Science of Language,'' in

R. Cohen and H. Levine, eds., Maimonides and the Sciences [Boston Studies in the
History and Philosophy of Science] (Dordrecht, Kluwer).
Stern, J. (ms.), ``A Little of `This' and a Little of `That'.''
Strauss, L. (1952), Persecution and the Art of Writing (Glencoe, Ill., The Free
Press).
Strawson, P. F. (1950), ``On Referring,'' Mind: 59: 320344.
Strawson, P. F. (1952), Introduction to Logical Theory (London, Metheun).
Sweetser, E. (1990), From Etymology to Pragmatics: Metaphorical and Cultural
Aspects of Semantic Structure (Cambridge, Cambridge University Press).
Sweetser, E. (1992), ``English Metaphors for Language: Motivations, Con-
ventions, and Creativity,'' Poetics Today 13: 705736.
Swinburne, R. (1992), Revelation: From Metaphor to Analogy (Oxford, Oxford
University Press).
Taschek, W. (1987), ``Content, Character, and Cognitive Signicance,'' Philo-
sophical Studies 52: 161189.
Taylor, B. (1980), ``Truth Theory for Indexical Languages,'' in M. Platts, ed.,
Reference, Truth, and Reality (London, Routledge): 182198.
Thompson, A. and Thompson, J. O. (1987), Shakespeare, Meaning, and Metaphor
(Iowa City, University of Iowa Press).
Tirrell, L. (1989), ``Extending: The Structure of Metaphor,'' Nous 23: 1734.
Tirrell, L. (1991), ``Reductive and Nonreductive Simile Theories of Metaphor,''
Journal of Philosophy 88: 337358.
Tormey, A. (1983), ``Metaphors and Counterfactuals,'' in J. Fisher, ed., Essays on
Aesthetics: Perspectives on the Work of Monroe C. Beardsley (Philadelphia, Tem-
ple University Press): 235246.
Torreano, L. A. (1997), Understanding Metaphorical Use of Verbs (Ph.D. disser-
tation, Princeton University).
Turbayne, C. (1970), The Myth of Metaphor (Columbia, S.C., University of South
Carolina Press).
Turner, M. (1987), Death Is the Mother of Beauty (Chicago, University of
Chicago Press).
Tversky, A. (1977), ``Features of Similarity,'' Psychological Review 84: 322
352.
van Fraassen, B. (1978), ``Presupposition, Implication, and Self-reference,'' Jour-
nal of Philosophy 65: 136151.
Vickers, B. (1988), ``Rhetoric and Poetics,'' in Quentin Skinner et al., eds., Cam-
bridge History of Renaissance Philosophy (Cambridge, Cambridge University
Press).
Walton, K. (1993), ``Metaphor and Prop-Oriented Make-Belief,'' European Jour-
nal of Philosophy 1: 3957.
378 References

Washington, C. (1992), ``The Identity Theory of Quotation,'' Journal of Philoso-

phy 89: 551581.
Weinstein, S. (1974), ``Truth and Demonstratives, '' Nous 8: 179184.
Wettstein, H. (1986), ``Has Semantics Rested on a Mistake?'' Journal of Philoso-
phy 83: 185209.
White, R. W. (1996), The Structure of Metaphor: The Way the Language of Meta-
phor Works (Oxford, Basil Blackwell).
White, S. (1982), ``The Language of Thought,'' Pacic Philosophical Quarterly 63:
347365.
Williams, E. S. (1977), ``Discourse and Logical Form,'' Linguistic Inquiry 8: 101
139.
Williams, J. (1976), ``Synaesthetic Adjectives: A Possible Law of Semantic
Change,'' Language 52: 461478.
Winner, E. (1988), The Point of Words: Children's Understanding of Metaphor and
Irony (Cambridge, Mass., Harvard University Press).
Winner, E. and Gardner, H. (1978), ``The Development of Metaphoric Compe-
tence: Implications for Humanistic Disciplines,'' Critical Inquiry 5: 123141.
Winner, E. and Gardner, H. (1993), ``Metaphor and Irony: Two Levels of
Understanding,'' in Ortony (1993): 425445.
Wittgenstein, L. (1980), Culture and Value, trans. P. Winch (Chicago, University
of Chicago Press).
Wollheim, R. (1980), Art and Its Objects, second edition (Cambridge, Cambridge
University Press).
Zangwill, N. (1991), ``Metaphor and Realism in Aesthetics,'' Journal of Aesthetics
and Art Criticism 49: 5762.
Index

Ambiguity, 73 Carey, J., 11

and fallacy of equivocation, 73, 217
idiosyncratic lexical, 7374
metaphorical/literal, 4, 74, 134, 202
and metaphorical recognition, 6, 202
of scope of `Mthat' operator, 129, 134
136
systematic lexical, 74
American Sign Language (ASL), 189191
Analog/digital distinction, 290293. See
also Pictorialness
Analytic/synthetic distinction, 175, 241, 243,
262
Analytic truths, metaphors and, 288
Anaphora, 244245, 174
and constraints on meaning, 6466
verb phrase, 6970, 215217
Aristotle, 7, 10, 27, 29, 152, 173, 229, 262,
275276
Assertion
by metaphor, 24, 118, 120121, 127, 147,
204, 230, 282283, 304
v. presupposition, 114, 226227
reports of, 209214
and resistibility, 286288
v. what is said, 142143
Asymmetrical directionality, 148151. See
also Similarity
Augustine, 305
Austin, J. L., 137
Beardsley, M., 73
Black, M., 12, 2122, 242, 282283
on creative metaphors, 120121, 148,
150
and cognitive autonomy, 264265
on system of associated commonplaces,
107109
Booth, W., 156
Borges, J., 27, 29
Brooks, C., 264
Burge, T., 188, 284
Catachresis, 181, 189190, 266
Cavell, S., 268270, 294
Character, 16, 82, 85. See also Content;
Meaning
-assignment stage, 82
``character-istic'' information, 20, 85 103,
279, 295297
as constraint on content, 88, 92, 105, 107,
255
of dthat-descriptions, 100
of indexicals, 8892
knowledge of, 253
limitations of, 262, 292, 297
linguistic meaning as, 16, 8889
as mode of presentation, 95100, 103, 273,
297
(non)constant, denition of, 16, 82
(non)constant, de jure/de facto, 101
of proper demonstratives, 94
underdetermines content, 18, 256
and logical validity, 8889
Character, metaphorical. See Metaphorical
character
Chomsky, N., 30
Class-inclusion statements, 184
Cognitive autonomy thesis, 264, 299
Cohen, G. A., 112
Cohen, T., 192, 202, 305
on irony, 233
on twice-true metaphors, 239 (see also
Metaphors, twice-true)
Communication, models of, 3941, 119
120, 127128
Comparison theory. See Similarity;
Simile
Complex gurative statements, 235236
Compositionality, 25, 47, 4951, 7273,
200201, 221, 307
Concealment
through demonstratives, 196
through metaphors, 138, 195
380
Conditional statements, pragmatic analysis
of, 207
Content (propositional), 1522, 35, 44,
6671, 77, 8083, 93, 104. See also
Propositions
and circumstances of evaluation, 82
and direct reference, 8398
possible worlds semantics analysis of, 82
of variables, 84
as what is said/asserted, 2, 142143
Context, 82, 92. See also Context-
dependence; Occurrences of sentences v.
utterances
actual context constraint, 71, 210212
as context set of presuppositions, 313
113115,119120, 145
linguistic v. extralinguistic, 22, 110111,
139, 243245
as semantic abstraction, 9192, 136
as total speech situation, 91
Context-dependence. See also Context;
Demonstratives; Indexicals (proper)
as extralinguistic use, 1314
as irregular and unpredictable, 1011
postsemantic, 43, 62, 83, 137
presemantic, 42, 62, 82, 308309, 315
semantic, 62, 83
and semantic constraints, 14, 257, 313
and semantics, 6263
systematic regularities of, 1113, 70,
105, 138139, 156176, 183, 257, 268
269
Continuum, metaphorical-literal, 316317
Contrast set, 169, 242, 246247
and exemplication-schema, 155156
Counterfactuals, metaphors in, 207
Crimmins, M., 109. See also Normal
notions
Culler, J., 6
D'Avanzo, M. L., 111
Davidson, D., 10, 19, 3536, 3967, 68, 72,
74, 133, 165, 197, 214, 221, 249, 299
on autonomy of meaning, 43
causal theory of metaphor, 4549, 5155
on demonstratives, 6163
denies metaphorical content, 265, 268269,
270271, 285286, 289, 294, 299
denies metaphorical meaning, 3845
on endlessness of metaphorical interpreta-
tion, 268270, 289
on rst meaning, 4142
on jokes, 46
on language, 4041
on literal meaning, 3944, 4849, 221, 303
on malapropisms, 5558
on meaning as truth-conditions, 3940,
6061
Index
on passing and prior theories of inter-
pretation, 4041, 5760
on pictures and metaphors, 46, 285286,
289, 294
on referential denite description, 5558
on semantics, 35, 40, 72, 270
on understanding, 3940, 5860, 6768
Davies, M., 283
Deconstructionism, 218
Decoration, metaphor as, 178, 262264, 299
and noncognitivism, 263264
De dicto
as fully conceptualized, 188189,191192
as literal meaning, 318
Denite description, 78
attributive use, 55
referential use, 5558, 226
Demonstrations, 78, 93
appearance of, 99 (see also Pictorialness)
Fregean theory of, 93, 97, 100
as gesture of pointing, 94
as presentations, 94, 97100, 256
Demonstratives, 9, 1419, 45, 77, 267, 295,
299, 301302. See also Demonstrations;
De re reference (expression); Dthat
descriptions; Indexicals (proper);
Reference, direct
analogy with metaphors, 1519, 45, 60, 71,
75, 88, 105, 106, 125, 132, 196, 204, 209
210, 240, 243, 307, 309
character of, 94, 206207, 267, 313
complete v. incomplete, 78, 88, 9394, 191,
199, 256, 273
Davidson on, 6164, 72
demonstrative interpretations v. expres-
sions, 77, 103104,106
as directly referential, 8384, 88
dierences from indexicals, 7778
dierences from metaphors, 253257,
279
and dthat-descriptions, 9394, 199
Frege's puzzle for, 95100, 103, 272273
counterevidence to ``Old'' theory, 81
and perspectivalism, 80
predicative, 8588, 187189, 206207, 256,
284
as rigid designators, 79
simple v. complex, 78
Denotation, 7980, 8384, 106, 188189,
200, 208, 226, 228
and dthat-descriptions, 100103, 199
and indexicals, 8889, 92
as indirect v. direct reference, 84, 8788
(not) thoroughly nondenotational, 8485,
89, 93, 101
as satisfaction or t, 89, 188
Dependence, metaphorical-literal. See
Literal meaning, metaphor depends on
Index
De re reference (expression), 188189, 191
2, 316
and metaphors of exemplication, 190,
194
and irreducible metaphors, 192194
and knowledge by metaphor, 192, 284
Deviance, grammatical (semantic anomaly)
and semantic theories of metaphor, 238
239, 243
and knowledge that metaphor, 36, 133
134, 243246
and surprise by metaphor, 276
Direct reference. See Reference, direct
Donnellan, K., 45, 5556, 93, 226227
Double-function terms, 310
Dthat-descriptions, 19, 77, 93104, 228,
256. See also Character
v. `Mthat[]', 199201, 255257
not thoroughly nondenotational, 93, 101
and referential denite description, 93,
226
as schematic rule, 100101, 197, 199201
syntax of, 94, 101104
Empson, W., 267268
Endlessness of metaphorical interpretation,
249250, 267271. See also Indetermi-
nacy of metaphor
and pictorialness of metaphors, 289294
Evaluation, 68, 136
appropriateness-, 127132, 136138
circumstances of, 8182, 113
v. interpretation, 205209, 214
of theory, 2627
truth-, 116, 137, 237 (see also Truth-
valuedness of metaphors)
Evans, G., 191
Exemplication, 20, 146, 154156, 184187,
191, 194195, 242, 279, 317. See also De
re reference (expression); Goodman, N.
and de re, 189191
-metaphors, 111, 140, 156168, 185 (see
also Normal notions; Stereotypical fea-
tures)
as mode of reference, 154
and recognition, 192
related to moral, allegorical, analogical
signication, 305
and sampling, 154155
schema-relativity of, 155164, 168169
Finite dierentiability, 290291
Fogelin, R., 34, 219221, 223224, 230
231, 234235
Formalization, limits of, 103104, 262, 293,
307
Frege, G., 80, 87, 154, 188, 211, 265, 272,
274, 279, 281, 293
381
Fregean theory of demonstrations, 97100
Puzzle of identity, 78, 95, 103, 272
`Good', 219
Glucksberg, S., 152153, 183184, 187,
189191, 230
Goldblum, N., 4749, 59, 222
Goodman, N., 108, 242, 276
on analog/digital, 290291
on exemplication, 154156
on family (schema)-relativity of meta-
phorical transfer, 162169 (see also
Metaphors, individuation of; Networks)
on simile and metaphor, 230,
on symbolic skills, 302
Grammar, metaphor and 133134, 240
Grice, H. P., 34, 3637, 233234
Gricean maxims, 127128
Gruber, J. S., 180
Haugeland, J., 290292
Higginbotham, J., 278
Hobbes, J., 263
Housman, A. E., 263
Hume, D., 112
Husserl, E., 94
Hyperbole, 237
`I', 8889
Ibn Ezra, M., 262263
Identication of metaphor. See Knowledge
that metaphor (``recognition question'')
I-gures, 237
Imagery, 289
Incompetence, metaphorical, 201205
Inconsistency, rule of, 128, 140141
Indeterminacy of metaphor, 160, 249252,
255257. See also Endlessness of meta-
phorical interpretation
Indexicals (proper), 14, 6164, 66, 7778,
8183, 96, 104, 107, 113, 219220. See
also Demonstratives
and actual context constraint, 210212
character of, 8890, 269
disanalogies with metaphors, 253257
``essential,'' 295296
parametric, 8893, 101, 205206
and pragmatically necessary truths, 8889,
288
systematic nature of, 211, 278279
thoroughly nondenotational, 8992
Indirect speech acts, 33, 222225, 232
233
Inductive network, 169, 174176
Intended v. possible interpretation, 3,
202. See also Teleological use of
metaphors
Irony, 36, 223, 225, 229, 232238, 248
382
Jackendo, R., 180183. See also Lako,
G.
Johnson, M., 176
Kant, I., 7, 10, 112
Kaplan, D., 14, 19, 35, 61, 77102, 187, 272,
295
Kenner, H., 173
Keysar, B., 152153, 183184, 187, 189
191, 230
Kittay, E. F., 21
on grammatical deviance, 243246
``perspectival theory,'' 242248
on semantic elds and metaphor, 246248
on syntagmatic relations, 169171
Knowledge by metaphor, 79, 145, 115
117, 145146, 260. See also Metaphorical
character; Metaphors, essential; Perspec-
tive, cognitive; Surprise
de re knowledge, 187196
and endlessness of metaphorical interpre-
tation, 271, 289294
individuated by character, 200201
and knowledge of metaphor, 8, 200201,
261, 266
and literal paraphrasability, 257, 262267
manifest in extended metaphors, 241, 280
281
as mode of presentation, 272274
nonpropositional, 285286, 289294, 299
as perspective or way of thinking, 277
281
and propositional attitudes, 259260, 295
299
and seeing-as, 288
Knowledge of metaphor, 79, 1516, 115,
197205, 243, 248 255, 257. See also
Metaphorical character; `Mthat[]';
Presupposition
and knowledge by metaphor, 79, 200
201, 261, 266
and pragmatics, 145146
as semantic knowledge, 1819, 6768,
105106, 132143
Knowledge that metaphor (``recognition
question''), 36, 37, 133136, 201202,
244246, 316. See also Deviance, gram-
matical (semantic anomaly); Parallel vs.
serial processing
Kripke, S., 267
Lako, G., 2122, 108, 169, 301, 303. See
also Jackendo, R.; Glucksberg, S.;
Keysar, B.
``contemporary'' or conceptual theory of
metaphor, 176, 187
Lehrer, A., 169
Index
Literal meaning, 2324, 39, 303, 316318
and `actual', 304
as atomistic, 317
autonomy of, 6367, 6971
and compositionality, 49, 60, 303, 306307
contains metaphorical meaning, 240242
and context-(in)dependence, 308309,
316319
as extension, 50
as rst meaning, 4045, 3943 (see also
Davidson, D.)
and fully conceptualized understanding,
191, 318
historical notions of, 305306
as intension, 5051
metaphor depends on, 17, 23, 4951, 59
60, 157, 221222, 240241, 311312, 319
metaphor ``has'' its, 17, 5859, 167, 220
221
as semantic interpretation, 23, 306
sensus literatus, 305
Literal vehicle, 24, 50, 147, 156157, 161,
167168, 221222, 269
Living metaphors. See Metaphors, dead v.
living
Locke, J., 263
Logical positivists, 264
Mailer, N., 156
Maimonides, M., 195
Malapropisms, 5558
Margalit, A., 4749, 52, 59, 222
m-associated features, 113115, 118, 126
128, 130132, 139, 141, 146147, 156,
167168, 198200, 202204, 221, 252,
254, 314. See also Properties, meta-
phorically relevant)
Meaning. See also Ambiguity; Character;
Content; Literal meaning; Metaphorical
meaning
adequacy conditions for, 45, 61, 78
as constraint, 13, 23, 33, 6371, 88, 197,
214217, 251252, 255
rst, 4143
sentence-, 37, 222
speaker's, 33, 36, 222223, 263
Meiosis, 229, 237
Metaphorical character, 16, 105, 159. See
also Metaphorical meaning
and actual context constraint, 71, 205207
and character-istic information, 20, 22, 85,
88, 103, 116, 166, 168, 189, 197, 203, 228,
259263, 269, 273274, 278280, 298,
317
as constraint on content, 105, 107, 133,
138, 141, 143, 145, 169, 176, 197, 243, 257
as ``directions,'' 269
Index
hybrid nature of, 106, 197201
limitations of, 262, 293, 297
and metaphorical incompetence, 203205
as metaphorical meaning, 1617,197, 214
222
as mode of presentation of content, 273
274, 280281
rule of character for `Mthat[]', 115
schematic, 106, 198199
semantic status of, 132143
Metaphorical expression, 106. See also
`Mthat[]'
Metaphorical-literal dependence. See Literal
meaning, metaphor depends on
Metaphorical meaning, 1617, 3645, 67
75, 197, 205222, 225, 32233, 238, 240,
242, 273. See also Metaphorical character
Metaphors
acquired, 314315 (see also Metaphors,
dead v. living)
biblical, 262263
children's metaphors, 310, 315
conceptual (see Lako, G.)
conventional, 176179 (see also Lako,
G.)
core v. peripheral, 3031
dead v. living, 2531, 236, 257, 309316
``essential'', 298299
extended, 174, 280281
grounds of, 112, 145 (see also Exemplica-
tion; Stereotypical features; Normal
notions)
individuation of, 164169, 317 (see also
Unit of metaphor)
irreducible, 192194 (see also Theological
discourse, metaphors in)
in Islamic and Jewish Aristotelianism, 195
mixed, 128130
nominative, 225229
and other gures, 23, 229, 232233, 235
238
pictorial, 301303
poetic, 2931, 160, 178, 271
predicative, 147, 228229
root-inized, 312315
routinized, 126, 311313, 315
schema-relativity of (see Schema-relativity
of metaphor)
in science, 269270
of transfer v. extension, 27
and truth, 2425, 218, 237, 248249, 266,
304 (see also Truth-valuedness of meta-
phors)
twice-true, 4, 163, 239
`Metaphor', wide and narrow senses of, 23.
See also Metaphors, and other gures
Metonymy, 229, 237
383
M-gures, 237
Moran, R., 286288
Morris, C., 154
`Mthat[]', 105106
knowledge of skill-like operator, 106, 200
201, 252, 252
metaphor set, 133134, 201202, 239
modeled on `dthat', 106, 198201
parametric, 115, 200, 221
predicate demonstrative-like, 107, 115, 198,
256
schematic rule of character, 115117, 198
syntax of, 198, 200, 221
thoroughly nondenotational, 200
Natural signs, 305
Networks, 166169. See also Exemplica-
tion, schema-relativity of; Inductive
network; Schema-relativity of metaphor;
Thematic network (scheme)
Noise, 54
Normal notions, 109110, 126, 131, 153,
302, 312. See also Salience, and exempli-
cation-metaphors; Properties, metaphor-
ically relevant; Stereotypical features
Nozick, R., 112
Occurrences of sentences v. utterances, 14,
91
Ornament, metaphor as. See Decoration,
metaphor as
Ortony, A., 1512
Overstatement, 237
Parallel vs. serial processing, 56, 133134
Paraphrasability, literal, 8, 257, 261, 262
267, 269274, 280, 292294. See also
Cognitive autonomy thesis; Endlessness
of metaphorical interpretation; Knowl-
edge by metaphor
and Frege's puzzle, 273274
Pascal, B., 195
Peirce, C. S., 154
Perry, J., 295296,
Perspectivalism, 80, 281
Perspective, cognitive, 80, 93, 9596, 272,
276. See also Perspectivalism; Schema-
relativity of metaphor
of indexical language, 278279
metaphorical, 277278, 280285, 289290,
292, 294, 298299
Pictorialness, 290292
in demonstrations, 99
and endlessness of metaphor, 286, 292294
in exemplication, 155156,
and irresistible force of metaphors, 286
287
384
Pictorialness (cont.)
of metaphors, 44, 161, 166, 190, 262, 289
290, 292294
and seeing-as, 285
Pragmatics
conceptions of, 3334, 38
and deviance, 3, 181, 239, 243244
of metaphor, 13, 1819, 113, 115116,
142143, 145146, 197, 226, 232233,
238, 246248, 251252, 254, 257
and semantics, 45, 75, 169, 176, 249, 251
252
Presupposition, 13, 1819, 22, 243, 250, 281
adopted v. native, 212214 (see also
Propositional attitudes, metaphors in)
A(ssertion)-set, 127, 140
v. assertion, 114, 143144, 228
cancellation of, 214
commitment to, 124, 130131
as common knowledge, 117118, 121
as contextual parameter for metaphor,
114117, 242
f(ilter)-, 139141, 159160
functions of, 117, 122, 126
identity of attitude of, 118
individual/shared context set of, 113, 119,
202, 314
making, 124
I(nterpretation)-set, 127, 138
and metaphorical incompetence, 202
metaphorically relevant v. conventional,
123132
and mixed metaphors, 130131
pragmatic notion of, 117, 122, 252
priority of, 118121
p(roductive)-, 139141, 159160
and reasonable inference, 123
semantic notion of, 117, 122
of sentence, 123
of speaker, 117
tests for, 114115
of utterance, 122123
Presupppositional inconsistency, rule of,
128, 130
Proper names, 228, 241, 267, 272, 293, 297
Properties
bare, 188, 193, 195, 284, 318
metaphorically relevant, 107115, 145,
147, 198, 200, 202, 221, 252, 302 (see also
Exemplication; m-associated features;
Normal notions; Stereotypical features)
semantics v. metaphysics of, 208
Propositional attitudes
and actual context constraint, 210212
demonstratives in, 211, 295299 (see also
Indexicals [proper], essential)
metaphors in, 209214, 259260, 295,
297299
Index
Propositions. See also Content (proposi-
tional)
general, 8587, 187188
purely (fully) conceptualized, 87, 188, 193,
296
referential, 8788, 188189
singular (Russellian), 85, 187188, 296
Putnam, H., 315
Quine, W. v. O., 10, 52, 141, 169, 222, 267,
280, 293
Quotation, 103, 199, 222
Recognition question. See Knowledge that
metaphor (``recognition question'')
Recognition v. recall, 191192
Redundancy, rule of, 128, 140141
Reference
direct, 8394, 100103, 187188, 199
(see also Denotation; De re reference
[expression]; Propositions)
Goodman on, 154155, 302 (see also
Exemplication)
indirect, 84, 188
``New'' theory of, 7883, 87
``Old'' theory of, 7883, 8789, 95
Reinhart, T., 174
Replication, token-, 98, 291292, 302
Reversiblity objection, 148150
Rhetorical tradition, 262263
Richards, I. A., 163164, 174, 264
Ricoeur, P., 266
Rigid (non-rigid) designation, 79, 8384, 86,
88, 9293, 98, 100, 102, 208. See also
Reference, direct
Romantics, 264
Rorty, R., 10, 5254
Rosch, E., 189
Rousseau, J.-J., 319
Rubinstein, B., 110
Russell, B., 15, 87, 188
Salience, 149154, 219220, 302
and exemplication-metaphors, 156
salience-imbalance, 1512
Sampling. See Exemplication
Schema-relativity of metaphor, 162164,
316317. See also Networks
and character, 174
and cognitive perspective, 279285
and individuation of metaphor, 165169
Searle, J., 34, 241
Fogelin on, 223224
on irony, 234
on literal meaning, 308309
on metaphor, 3639
Seeing-as, 20, 281282. See also Perspective,
cognitive; Pictorialness, of metaphors
Index 385

individuated by character, 289294 Thematic network (scheme), 169174, 187,

and assertion, 285288
as way of thinking of contents, 282285
Semantic competence, 910, 1516, 6769
169, 248253, 261, 302
Semantic elds, 170, 180, 182183, 242,
246248. See also Contrast set; Kittay,
E. F.
Semantics
compositional v. lexical, 7475
nitistic, 3435, 7275, 269270, 294
and semantic eld theory, 247248
as theory of sentences-in-contexts, 1415,
2526
as theory of form or constraints, 1719,
6168, 248253
Similarity, 145153. See also Salience
assertion by metaphor, 147, 230
asymmetry of judgments, 148, 150, 263
canons of, 219
as context of metaphor, 147, 149, 230
created by metaphor, 120121, 147148
geometrical v. feature matching models of,
148149
reversibility objection to, 148, 150151
``similarity thesis,'' 1467
and simile, 147, 152 (see also Simile)
Simile, 147148, 150152, 195, 198, 229
232, 237, 263. See also Similarity
asymmetry of, 148, 230
`like' as hedge, 232
metaphor as elliptic for, 147, 152153,
219220, 229232
Stalnaker, R., 113, 124
on pragmatic presupposition, 117118, 126
(see also Presupposition)
Stereotypical features, 109, 118, 126, 131,
302, 312. See also Exemplication;
m-associated features; Normal notions;
Properties, metaphorically relevant
Substitutivity, failure of, 50, 217. See also
Compositionality
Surprise, 262, 274. See also Deviance,
grammatical (semantic anomaly); Tension
of metaphor
as cognitive emotion, 275276
Fregean analysis of, 274275
as improbability of token, 275
as unpredictability, 275
Synaesthetic adjectives, 310
Synechdoche, 229, 237
242. See also Networks
and syntagmatic relations, 170, 242
thematic roles, 170, 180, 186
verb classes, 171
Thematic relations hypothesis, 180, 182
Theological discourse, metaphors in, 192
194
Tirrell, L., 169, 174
Transfer
Goodman on, 162164, 302 (see also
Goodman, N.; Metaphors, individuation
of )
metaphors of, 27
Truth-valuedness of metaphors, 2425,
136137,248251. See also Indeterminacy
of metaphor; Endlessness of metaphorical
interpretation
Tversky, A., 148153, 159. See also
Similarity; Salience
Understanding, 23, 221222
psychological processes of, 2, 184
and truth-conditions, 3940, 4749, 5960,
6364, 6768, 78
Understatement, 229, 237
Unit of metaphor, 2123. See also Meta-
phors, individuation of
Use theory. See Pragmatics
Vickers, B., 263
White, R. W., 22, 4749, 59, 169, 222
on metaphorical ambiguity, 135136
Wittgenstein, L., 33, 162, 165, 286

Teleological use of metaphors, 3, 227, 269

270
Tension of metaphor, 238, 243. See also
Deviance, grammatical (semantic
anomaly); Surprise